Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * hugetlbpage-backed filesystem. Based on ramfs. | |
3 | * | |
6d49e352 | 4 | * Nadia Yvette Chambers, 2002 |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 2002 Linus Torvalds. | |
3e89e1c5 | 7 | * License: GPL |
1da177e4 LT |
8 | */ |
9 | ||
9b857d26 AM |
10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
11 | ||
1da177e4 LT |
12 | #include <linux/thread_info.h> |
13 | #include <asm/current.h> | |
14 | #include <linux/sched.h> /* remove ASAP */ | |
70c3547e | 15 | #include <linux/falloc.h> |
1da177e4 LT |
16 | #include <linux/fs.h> |
17 | #include <linux/mount.h> | |
18 | #include <linux/file.h> | |
e73a75fa | 19 | #include <linux/kernel.h> |
1da177e4 LT |
20 | #include <linux/writeback.h> |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/highmem.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/string.h> | |
16f7e0fe | 25 | #include <linux/capability.h> |
e73a75fa | 26 | #include <linux/ctype.h> |
1da177e4 LT |
27 | #include <linux/backing-dev.h> |
28 | #include <linux/hugetlb.h> | |
29 | #include <linux/pagevec.h> | |
e73a75fa | 30 | #include <linux/parser.h> |
036e0856 | 31 | #include <linux/mman.h> |
1da177e4 LT |
32 | #include <linux/slab.h> |
33 | #include <linux/dnotify.h> | |
34 | #include <linux/statfs.h> | |
35 | #include <linux/security.h> | |
1fd7317d | 36 | #include <linux/magic.h> |
290408d4 | 37 | #include <linux/migrate.h> |
34d0640e | 38 | #include <linux/uio.h> |
1da177e4 LT |
39 | |
40 | #include <asm/uaccess.h> | |
41 | ||
ee9b6d61 | 42 | static const struct super_operations hugetlbfs_ops; |
f5e54d6e | 43 | static const struct address_space_operations hugetlbfs_aops; |
4b6f5d20 | 44 | const struct file_operations hugetlbfs_file_operations; |
92e1d5be AV |
45 | static const struct inode_operations hugetlbfs_dir_inode_operations; |
46 | static const struct inode_operations hugetlbfs_inode_operations; | |
1da177e4 | 47 | |
a1d776ee | 48 | struct hugetlbfs_config { |
a0eb3a05 EB |
49 | kuid_t uid; |
50 | kgid_t gid; | |
a1d776ee | 51 | umode_t mode; |
7ca02d0a | 52 | long max_hpages; |
a1d776ee DG |
53 | long nr_inodes; |
54 | struct hstate *hstate; | |
7ca02d0a | 55 | long min_hpages; |
a1d776ee DG |
56 | }; |
57 | ||
58 | struct hugetlbfs_inode_info { | |
59 | struct shared_policy policy; | |
60 | struct inode vfs_inode; | |
61 | }; | |
62 | ||
63 | static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode) | |
64 | { | |
65 | return container_of(inode, struct hugetlbfs_inode_info, vfs_inode); | |
66 | } | |
67 | ||
1da177e4 LT |
68 | int sysctl_hugetlb_shm_group; |
69 | ||
e73a75fa RD |
70 | enum { |
71 | Opt_size, Opt_nr_inodes, | |
72 | Opt_mode, Opt_uid, Opt_gid, | |
7ca02d0a | 73 | Opt_pagesize, Opt_min_size, |
e73a75fa RD |
74 | Opt_err, |
75 | }; | |
76 | ||
a447c093 | 77 | static const match_table_t tokens = { |
e73a75fa RD |
78 | {Opt_size, "size=%s"}, |
79 | {Opt_nr_inodes, "nr_inodes=%s"}, | |
80 | {Opt_mode, "mode=%o"}, | |
81 | {Opt_uid, "uid=%u"}, | |
82 | {Opt_gid, "gid=%u"}, | |
a137e1cc | 83 | {Opt_pagesize, "pagesize=%s"}, |
7ca02d0a | 84 | {Opt_min_size, "min_size=%s"}, |
e73a75fa RD |
85 | {Opt_err, NULL}, |
86 | }; | |
87 | ||
70c3547e MK |
88 | #ifdef CONFIG_NUMA |
89 | static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, | |
90 | struct inode *inode, pgoff_t index) | |
91 | { | |
92 | vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy, | |
93 | index); | |
94 | } | |
95 | ||
96 | static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) | |
97 | { | |
98 | mpol_cond_put(vma->vm_policy); | |
99 | } | |
100 | #else | |
101 | static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, | |
102 | struct inode *inode, pgoff_t index) | |
103 | { | |
104 | } | |
105 | ||
106 | static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) | |
107 | { | |
108 | } | |
109 | #endif | |
110 | ||
2e9b367c AL |
111 | static void huge_pagevec_release(struct pagevec *pvec) |
112 | { | |
113 | int i; | |
114 | ||
115 | for (i = 0; i < pagevec_count(pvec); ++i) | |
116 | put_page(pvec->pages[i]); | |
117 | ||
118 | pagevec_reinit(pvec); | |
119 | } | |
120 | ||
1da177e4 LT |
121 | static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) |
122 | { | |
496ad9aa | 123 | struct inode *inode = file_inode(file); |
1da177e4 LT |
124 | loff_t len, vma_len; |
125 | int ret; | |
a5516438 | 126 | struct hstate *h = hstate_file(file); |
1da177e4 | 127 | |
68589bc3 | 128 | /* |
dec4ad86 DG |
129 | * vma address alignment (but not the pgoff alignment) has |
130 | * already been checked by prepare_hugepage_range. If you add | |
131 | * any error returns here, do so after setting VM_HUGETLB, so | |
132 | * is_vm_hugetlb_page tests below unmap_region go the right | |
133 | * way when do_mmap_pgoff unwinds (may be important on powerpc | |
134 | * and ia64). | |
68589bc3 | 135 | */ |
a2fce914 | 136 | vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; |
68589bc3 | 137 | vma->vm_ops = &hugetlb_vm_ops; |
1da177e4 | 138 | |
2b37c35e | 139 | if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) |
dec4ad86 DG |
140 | return -EINVAL; |
141 | ||
1da177e4 LT |
142 | vma_len = (loff_t)(vma->vm_end - vma->vm_start); |
143 | ||
5955102c | 144 | inode_lock(inode); |
1da177e4 | 145 | file_accessed(file); |
1da177e4 LT |
146 | |
147 | ret = -ENOMEM; | |
148 | len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
1da177e4 | 149 | |
a1e78772 | 150 | if (hugetlb_reserve_pages(inode, |
a5516438 | 151 | vma->vm_pgoff >> huge_page_order(h), |
5a6fe125 MG |
152 | len >> huge_page_shift(h), vma, |
153 | vma->vm_flags)) | |
a43a8c39 | 154 | goto out; |
b45b5bd6 | 155 | |
4c887265 | 156 | ret = 0; |
b6174df5 | 157 | if (vma->vm_flags & VM_WRITE && inode->i_size < len) |
1da177e4 LT |
158 | inode->i_size = len; |
159 | out: | |
5955102c | 160 | inode_unlock(inode); |
1da177e4 LT |
161 | |
162 | return ret; | |
163 | } | |
164 | ||
165 | /* | |
508034a3 | 166 | * Called under down_write(mmap_sem). |
1da177e4 LT |
167 | */ |
168 | ||
d2ba27e8 | 169 | #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA |
1da177e4 LT |
170 | static unsigned long |
171 | hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | |
172 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
173 | { | |
174 | struct mm_struct *mm = current->mm; | |
175 | struct vm_area_struct *vma; | |
a5516438 | 176 | struct hstate *h = hstate_file(file); |
08659355 | 177 | struct vm_unmapped_area_info info; |
1da177e4 | 178 | |
a5516438 | 179 | if (len & ~huge_page_mask(h)) |
1da177e4 LT |
180 | return -EINVAL; |
181 | if (len > TASK_SIZE) | |
182 | return -ENOMEM; | |
183 | ||
036e0856 | 184 | if (flags & MAP_FIXED) { |
a5516438 | 185 | if (prepare_hugepage_range(file, addr, len)) |
036e0856 BH |
186 | return -EINVAL; |
187 | return addr; | |
188 | } | |
189 | ||
1da177e4 | 190 | if (addr) { |
a5516438 | 191 | addr = ALIGN(addr, huge_page_size(h)); |
1da177e4 LT |
192 | vma = find_vma(mm, addr); |
193 | if (TASK_SIZE - len >= addr && | |
194 | (!vma || addr + len <= vma->vm_start)) | |
195 | return addr; | |
196 | } | |
197 | ||
08659355 ML |
198 | info.flags = 0; |
199 | info.length = len; | |
200 | info.low_limit = TASK_UNMAPPED_BASE; | |
201 | info.high_limit = TASK_SIZE; | |
202 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); | |
203 | info.align_offset = 0; | |
204 | return vm_unmapped_area(&info); | |
1da177e4 LT |
205 | } |
206 | #endif | |
207 | ||
34d0640e | 208 | static size_t |
e63e1e5a | 209 | hugetlbfs_read_actor(struct page *page, unsigned long offset, |
34d0640e | 210 | struct iov_iter *to, unsigned long size) |
e63e1e5a | 211 | { |
34d0640e | 212 | size_t copied = 0; |
e63e1e5a BP |
213 | int i, chunksize; |
214 | ||
e63e1e5a | 215 | /* Find which 4k chunk and offset with in that chunk */ |
09cbfeaf KS |
216 | i = offset >> PAGE_SHIFT; |
217 | offset = offset & ~PAGE_MASK; | |
e63e1e5a BP |
218 | |
219 | while (size) { | |
34d0640e | 220 | size_t n; |
09cbfeaf | 221 | chunksize = PAGE_SIZE; |
e63e1e5a BP |
222 | if (offset) |
223 | chunksize -= offset; | |
224 | if (chunksize > size) | |
225 | chunksize = size; | |
34d0640e AV |
226 | n = copy_page_to_iter(&page[i], offset, chunksize, to); |
227 | copied += n; | |
228 | if (n != chunksize) | |
229 | return copied; | |
e63e1e5a BP |
230 | offset = 0; |
231 | size -= chunksize; | |
e63e1e5a BP |
232 | i++; |
233 | } | |
34d0640e | 234 | return copied; |
e63e1e5a BP |
235 | } |
236 | ||
237 | /* | |
238 | * Support for read() - Find the page attached to f_mapping and copy out the | |
239 | * data. Its *very* similar to do_generic_mapping_read(), we can't use that | |
ea1754a0 | 240 | * since it has PAGE_SIZE assumptions. |
e63e1e5a | 241 | */ |
34d0640e | 242 | static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) |
e63e1e5a | 243 | { |
34d0640e AV |
244 | struct file *file = iocb->ki_filp; |
245 | struct hstate *h = hstate_file(file); | |
246 | struct address_space *mapping = file->f_mapping; | |
e63e1e5a | 247 | struct inode *inode = mapping->host; |
34d0640e AV |
248 | unsigned long index = iocb->ki_pos >> huge_page_shift(h); |
249 | unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); | |
e63e1e5a BP |
250 | unsigned long end_index; |
251 | loff_t isize; | |
252 | ssize_t retval = 0; | |
253 | ||
34d0640e | 254 | while (iov_iter_count(to)) { |
e63e1e5a | 255 | struct page *page; |
34d0640e | 256 | size_t nr, copied; |
e63e1e5a BP |
257 | |
258 | /* nr is the maximum number of bytes to copy from this page */ | |
a5516438 | 259 | nr = huge_page_size(h); |
a05b0855 AK |
260 | isize = i_size_read(inode); |
261 | if (!isize) | |
34d0640e | 262 | break; |
a05b0855 | 263 | end_index = (isize - 1) >> huge_page_shift(h); |
34d0640e AV |
264 | if (index > end_index) |
265 | break; | |
266 | if (index == end_index) { | |
a5516438 | 267 | nr = ((isize - 1) & ~huge_page_mask(h)) + 1; |
a05b0855 | 268 | if (nr <= offset) |
34d0640e | 269 | break; |
e63e1e5a BP |
270 | } |
271 | nr = nr - offset; | |
272 | ||
273 | /* Find the page */ | |
a05b0855 | 274 | page = find_lock_page(mapping, index); |
e63e1e5a BP |
275 | if (unlikely(page == NULL)) { |
276 | /* | |
277 | * We have a HOLE, zero out the user-buffer for the | |
278 | * length of the hole or request. | |
279 | */ | |
34d0640e | 280 | copied = iov_iter_zero(nr, to); |
e63e1e5a | 281 | } else { |
a05b0855 AK |
282 | unlock_page(page); |
283 | ||
e63e1e5a BP |
284 | /* |
285 | * We have the page, copy it to user space buffer. | |
286 | */ | |
34d0640e | 287 | copied = hugetlbfs_read_actor(page, offset, to, nr); |
09cbfeaf | 288 | put_page(page); |
e63e1e5a | 289 | } |
34d0640e AV |
290 | offset += copied; |
291 | retval += copied; | |
292 | if (copied != nr && iov_iter_count(to)) { | |
293 | if (!retval) | |
294 | retval = -EFAULT; | |
295 | break; | |
e63e1e5a | 296 | } |
a5516438 AK |
297 | index += offset >> huge_page_shift(h); |
298 | offset &= ~huge_page_mask(h); | |
e63e1e5a | 299 | } |
34d0640e | 300 | iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; |
e63e1e5a BP |
301 | return retval; |
302 | } | |
303 | ||
800d15a5 NP |
304 | static int hugetlbfs_write_begin(struct file *file, |
305 | struct address_space *mapping, | |
306 | loff_t pos, unsigned len, unsigned flags, | |
307 | struct page **pagep, void **fsdata) | |
1da177e4 LT |
308 | { |
309 | return -EINVAL; | |
310 | } | |
311 | ||
800d15a5 NP |
312 | static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, |
313 | loff_t pos, unsigned len, unsigned copied, | |
314 | struct page *page, void *fsdata) | |
1da177e4 | 315 | { |
800d15a5 | 316 | BUG(); |
1da177e4 LT |
317 | return -EINVAL; |
318 | } | |
319 | ||
b5cec28d | 320 | static void remove_huge_page(struct page *page) |
1da177e4 | 321 | { |
b9ea2515 | 322 | ClearPageDirty(page); |
1da177e4 | 323 | ClearPageUptodate(page); |
bd65cb86 | 324 | delete_from_page_cache(page); |
1da177e4 LT |
325 | } |
326 | ||
4aae8d1c MK |
327 | static void |
328 | hugetlb_vmdelete_list(struct rb_root *root, pgoff_t start, pgoff_t end) | |
329 | { | |
330 | struct vm_area_struct *vma; | |
331 | ||
332 | /* | |
333 | * end == 0 indicates that the entire range after | |
334 | * start should be unmapped. | |
335 | */ | |
336 | vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { | |
337 | unsigned long v_offset; | |
338 | unsigned long v_end; | |
339 | ||
340 | /* | |
341 | * Can the expression below overflow on 32-bit arches? | |
342 | * No, because the interval tree returns us only those vmas | |
343 | * which overlap the truncated area starting at pgoff, | |
344 | * and no vma on a 32-bit arch can span beyond the 4GB. | |
345 | */ | |
346 | if (vma->vm_pgoff < start) | |
347 | v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; | |
348 | else | |
349 | v_offset = 0; | |
350 | ||
351 | if (!end) | |
352 | v_end = vma->vm_end; | |
353 | else { | |
354 | v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) | |
355 | + vma->vm_start; | |
356 | if (v_end > vma->vm_end) | |
357 | v_end = vma->vm_end; | |
358 | } | |
359 | ||
360 | unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, | |
361 | NULL); | |
362 | } | |
363 | } | |
b5cec28d MK |
364 | |
365 | /* | |
366 | * remove_inode_hugepages handles two distinct cases: truncation and hole | |
367 | * punch. There are subtle differences in operation for each case. | |
4aae8d1c | 368 | * |
b5cec28d MK |
369 | * truncation is indicated by end of range being LLONG_MAX |
370 | * In this case, we first scan the range and release found pages. | |
371 | * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv | |
1817889e MK |
372 | * maps and global counts. Page faults can not race with truncation |
373 | * in this routine. hugetlb_no_page() prevents page faults in the | |
374 | * truncated range. It checks i_size before allocation, and again after | |
375 | * with the page table lock for the page held. The same lock must be | |
376 | * acquired to unmap a page. | |
b5cec28d MK |
377 | * hole punch is indicated if end is not LLONG_MAX |
378 | * In the hole punch case we scan the range and release found pages. | |
379 | * Only when releasing a page is the associated region/reserv map | |
380 | * deleted. The region/reserv map for ranges without associated | |
1817889e MK |
381 | * pages are not modified. Page faults can race with hole punch. |
382 | * This is indicated if we find a mapped page. | |
b5cec28d MK |
383 | * Note: If the passed end of range value is beyond the end of file, but |
384 | * not LLONG_MAX this routine still performs a hole punch operation. | |
385 | */ | |
386 | static void remove_inode_hugepages(struct inode *inode, loff_t lstart, | |
387 | loff_t lend) | |
1da177e4 | 388 | { |
a5516438 | 389 | struct hstate *h = hstate_inode(inode); |
b45b5bd6 | 390 | struct address_space *mapping = &inode->i_data; |
a5516438 | 391 | const pgoff_t start = lstart >> huge_page_shift(h); |
b5cec28d MK |
392 | const pgoff_t end = lend >> huge_page_shift(h); |
393 | struct vm_area_struct pseudo_vma; | |
1da177e4 LT |
394 | struct pagevec pvec; |
395 | pgoff_t next; | |
a43a8c39 | 396 | int i, freed = 0; |
b5cec28d MK |
397 | long lookup_nr = PAGEVEC_SIZE; |
398 | bool truncate_op = (lend == LLONG_MAX); | |
1da177e4 | 399 | |
b5cec28d MK |
400 | memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); |
401 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); | |
1da177e4 LT |
402 | pagevec_init(&pvec, 0); |
403 | next = start; | |
b5cec28d MK |
404 | while (next < end) { |
405 | /* | |
1817889e | 406 | * Don't grab more pages than the number left in the range. |
b5cec28d MK |
407 | */ |
408 | if (end - next < lookup_nr) | |
409 | lookup_nr = end - next; | |
410 | ||
411 | /* | |
1817889e | 412 | * When no more pages are found, we are done. |
b5cec28d | 413 | */ |
1817889e MK |
414 | if (!pagevec_lookup(&pvec, mapping, next, lookup_nr)) |
415 | break; | |
1da177e4 LT |
416 | |
417 | for (i = 0; i < pagevec_count(&pvec); ++i) { | |
418 | struct page *page = pvec.pages[i]; | |
4aae8d1c | 419 | bool rsv_on_error; |
b5cec28d MK |
420 | u32 hash; |
421 | ||
1817889e MK |
422 | /* |
423 | * The page (index) could be beyond end. This is | |
424 | * only possible in the punch hole case as end is | |
425 | * max page offset in the truncate case. | |
426 | */ | |
427 | next = page->index; | |
428 | if (next >= end) | |
429 | break; | |
430 | ||
b5cec28d MK |
431 | hash = hugetlb_fault_mutex_hash(h, current->mm, |
432 | &pseudo_vma, | |
433 | mapping, next, 0); | |
434 | mutex_lock(&hugetlb_fault_mutex_table[hash]); | |
1da177e4 | 435 | |
4aae8d1c MK |
436 | /* |
437 | * If page is mapped, it was faulted in after being | |
438 | * unmapped in caller. Unmap (again) now after taking | |
439 | * the fault mutex. The mutex will prevent faults | |
440 | * until we finish removing the page. | |
441 | * | |
442 | * This race can only happen in the hole punch case. | |
443 | * Getting here in a truncate operation is a bug. | |
444 | */ | |
445 | if (unlikely(page_mapped(page))) { | |
1817889e | 446 | BUG_ON(truncate_op); |
4aae8d1c MK |
447 | |
448 | i_mmap_lock_write(mapping); | |
449 | hugetlb_vmdelete_list(&mapping->i_mmap, | |
450 | next * pages_per_huge_page(h), | |
451 | (next + 1) * pages_per_huge_page(h)); | |
452 | i_mmap_unlock_write(mapping); | |
453 | } | |
454 | ||
455 | lock_page(page); | |
456 | /* | |
457 | * We must free the huge page and remove from page | |
458 | * cache (remove_huge_page) BEFORE removing the | |
459 | * region/reserve map (hugetlb_unreserve_pages). In | |
460 | * rare out of memory conditions, removal of the | |
461 | * region/reserve map could fail. Before free'ing | |
462 | * the page, note PagePrivate which is used in case | |
463 | * of error. | |
464 | */ | |
465 | rsv_on_error = !PagePrivate(page); | |
466 | remove_huge_page(page); | |
467 | freed++; | |
468 | if (!truncate_op) { | |
469 | if (unlikely(hugetlb_unreserve_pages(inode, | |
470 | next, next + 1, 1))) | |
471 | hugetlb_fix_reserve_counts(inode, | |
472 | rsv_on_error); | |
b5cec28d MK |
473 | } |
474 | ||
1da177e4 | 475 | unlock_page(page); |
b5cec28d | 476 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
1da177e4 | 477 | } |
1817889e | 478 | ++next; |
1da177e4 | 479 | huge_pagevec_release(&pvec); |
1817889e | 480 | cond_resched(); |
1da177e4 | 481 | } |
b5cec28d MK |
482 | |
483 | if (truncate_op) | |
484 | (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); | |
1da177e4 LT |
485 | } |
486 | ||
2bbbda30 | 487 | static void hugetlbfs_evict_inode(struct inode *inode) |
1da177e4 | 488 | { |
9119a41e JK |
489 | struct resv_map *resv_map; |
490 | ||
b5cec28d | 491 | remove_inode_hugepages(inode, 0, LLONG_MAX); |
9119a41e JK |
492 | resv_map = (struct resv_map *)inode->i_mapping->private_data; |
493 | /* root inode doesn't have the resv_map, so we should check it */ | |
494 | if (resv_map) | |
495 | resv_map_release(&resv_map->refs); | |
dbd5768f | 496 | clear_inode(inode); |
149f4211 CH |
497 | } |
498 | ||
1da177e4 LT |
499 | static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) |
500 | { | |
856fc295 | 501 | pgoff_t pgoff; |
1da177e4 | 502 | struct address_space *mapping = inode->i_mapping; |
a5516438 | 503 | struct hstate *h = hstate_inode(inode); |
1da177e4 | 504 | |
a5516438 | 505 | BUG_ON(offset & ~huge_page_mask(h)); |
856fc295 | 506 | pgoff = offset >> PAGE_SHIFT; |
1da177e4 | 507 | |
7aa91e10 | 508 | i_size_write(inode, offset); |
83cde9e8 | 509 | i_mmap_lock_write(mapping); |
6b2dbba8 | 510 | if (!RB_EMPTY_ROOT(&mapping->i_mmap)) |
1bfad99a | 511 | hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); |
83cde9e8 | 512 | i_mmap_unlock_write(mapping); |
b5cec28d | 513 | remove_inode_hugepages(inode, offset, LLONG_MAX); |
1da177e4 LT |
514 | return 0; |
515 | } | |
516 | ||
70c3547e MK |
517 | static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
518 | { | |
519 | struct hstate *h = hstate_inode(inode); | |
520 | loff_t hpage_size = huge_page_size(h); | |
521 | loff_t hole_start, hole_end; | |
522 | ||
523 | /* | |
524 | * For hole punch round up the beginning offset of the hole and | |
525 | * round down the end. | |
526 | */ | |
527 | hole_start = round_up(offset, hpage_size); | |
528 | hole_end = round_down(offset + len, hpage_size); | |
529 | ||
530 | if (hole_end > hole_start) { | |
531 | struct address_space *mapping = inode->i_mapping; | |
532 | ||
5955102c | 533 | inode_lock(inode); |
70c3547e MK |
534 | i_mmap_lock_write(mapping); |
535 | if (!RB_EMPTY_ROOT(&mapping->i_mmap)) | |
536 | hugetlb_vmdelete_list(&mapping->i_mmap, | |
537 | hole_start >> PAGE_SHIFT, | |
538 | hole_end >> PAGE_SHIFT); | |
539 | i_mmap_unlock_write(mapping); | |
540 | remove_inode_hugepages(inode, hole_start, hole_end); | |
5955102c | 541 | inode_unlock(inode); |
70c3547e MK |
542 | } |
543 | ||
544 | return 0; | |
545 | } | |
546 | ||
547 | static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, | |
548 | loff_t len) | |
549 | { | |
550 | struct inode *inode = file_inode(file); | |
551 | struct address_space *mapping = inode->i_mapping; | |
552 | struct hstate *h = hstate_inode(inode); | |
553 | struct vm_area_struct pseudo_vma; | |
554 | struct mm_struct *mm = current->mm; | |
555 | loff_t hpage_size = huge_page_size(h); | |
556 | unsigned long hpage_shift = huge_page_shift(h); | |
557 | pgoff_t start, index, end; | |
558 | int error; | |
559 | u32 hash; | |
560 | ||
561 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
562 | return -EOPNOTSUPP; | |
563 | ||
564 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
565 | return hugetlbfs_punch_hole(inode, offset, len); | |
566 | ||
567 | /* | |
568 | * Default preallocate case. | |
569 | * For this range, start is rounded down and end is rounded up | |
570 | * as well as being converted to page offsets. | |
571 | */ | |
572 | start = offset >> hpage_shift; | |
573 | end = (offset + len + hpage_size - 1) >> hpage_shift; | |
574 | ||
5955102c | 575 | inode_lock(inode); |
70c3547e MK |
576 | |
577 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
578 | error = inode_newsize_ok(inode, offset + len); | |
579 | if (error) | |
580 | goto out; | |
581 | ||
582 | /* | |
583 | * Initialize a pseudo vma as this is required by the huge page | |
584 | * allocation routines. If NUMA is configured, use page index | |
585 | * as input to create an allocation policy. | |
586 | */ | |
587 | memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); | |
588 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); | |
589 | pseudo_vma.vm_file = file; | |
590 | ||
591 | for (index = start; index < end; index++) { | |
592 | /* | |
593 | * This is supposed to be the vaddr where the page is being | |
594 | * faulted in, but we have no vaddr here. | |
595 | */ | |
596 | struct page *page; | |
597 | unsigned long addr; | |
598 | int avoid_reserve = 0; | |
599 | ||
600 | cond_resched(); | |
601 | ||
602 | /* | |
603 | * fallocate(2) manpage permits EINTR; we may have been | |
604 | * interrupted because we are using up too much memory. | |
605 | */ | |
606 | if (signal_pending(current)) { | |
607 | error = -EINTR; | |
608 | break; | |
609 | } | |
610 | ||
611 | /* Set numa allocation policy based on index */ | |
612 | hugetlb_set_vma_policy(&pseudo_vma, inode, index); | |
613 | ||
614 | /* addr is the offset within the file (zero based) */ | |
615 | addr = index * hpage_size; | |
616 | ||
617 | /* mutex taken here, fault path and hole punch */ | |
618 | hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping, | |
619 | index, addr); | |
620 | mutex_lock(&hugetlb_fault_mutex_table[hash]); | |
621 | ||
622 | /* See if already present in mapping to avoid alloc/free */ | |
623 | page = find_get_page(mapping, index); | |
624 | if (page) { | |
625 | put_page(page); | |
626 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
627 | hugetlb_drop_vma_policy(&pseudo_vma); | |
628 | continue; | |
629 | } | |
630 | ||
631 | /* Allocate page and add to page cache */ | |
632 | page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); | |
633 | hugetlb_drop_vma_policy(&pseudo_vma); | |
634 | if (IS_ERR(page)) { | |
635 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
636 | error = PTR_ERR(page); | |
637 | goto out; | |
638 | } | |
639 | clear_huge_page(page, addr, pages_per_huge_page(h)); | |
640 | __SetPageUptodate(page); | |
641 | error = huge_add_to_page_cache(page, mapping, index); | |
642 | if (unlikely(error)) { | |
643 | put_page(page); | |
644 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
645 | goto out; | |
646 | } | |
647 | ||
648 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
649 | ||
650 | /* | |
651 | * page_put due to reference from alloc_huge_page() | |
652 | * unlock_page because locked by add_to_page_cache() | |
653 | */ | |
654 | put_page(page); | |
655 | unlock_page(page); | |
656 | } | |
657 | ||
658 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
659 | i_size_write(inode, offset + len); | |
660 | inode->i_ctime = CURRENT_TIME; | |
70c3547e | 661 | out: |
5955102c | 662 | inode_unlock(inode); |
70c3547e MK |
663 | return error; |
664 | } | |
665 | ||
1da177e4 LT |
666 | static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) |
667 | { | |
2b0143b5 | 668 | struct inode *inode = d_inode(dentry); |
a5516438 | 669 | struct hstate *h = hstate_inode(inode); |
1da177e4 LT |
670 | int error; |
671 | unsigned int ia_valid = attr->ia_valid; | |
672 | ||
673 | BUG_ON(!inode); | |
674 | ||
675 | error = inode_change_ok(inode, attr); | |
676 | if (error) | |
1025774c | 677 | return error; |
1da177e4 LT |
678 | |
679 | if (ia_valid & ATTR_SIZE) { | |
680 | error = -EINVAL; | |
1025774c CH |
681 | if (attr->ia_size & ~huge_page_mask(h)) |
682 | return -EINVAL; | |
683 | error = hugetlb_vmtruncate(inode, attr->ia_size); | |
1da177e4 | 684 | if (error) |
1025774c | 685 | return error; |
1da177e4 | 686 | } |
1025774c CH |
687 | |
688 | setattr_copy(inode, attr); | |
689 | mark_inode_dirty(inode); | |
690 | return 0; | |
1da177e4 LT |
691 | } |
692 | ||
7d54fa64 AV |
693 | static struct inode *hugetlbfs_get_root(struct super_block *sb, |
694 | struct hugetlbfs_config *config) | |
1da177e4 LT |
695 | { |
696 | struct inode *inode; | |
1da177e4 LT |
697 | |
698 | inode = new_inode(sb); | |
699 | if (inode) { | |
700 | struct hugetlbfs_inode_info *info; | |
85fe4025 | 701 | inode->i_ino = get_next_ino(); |
7d54fa64 AV |
702 | inode->i_mode = S_IFDIR | config->mode; |
703 | inode->i_uid = config->uid; | |
704 | inode->i_gid = config->gid; | |
705 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
706 | info = HUGETLBFS_I(inode); | |
707 | mpol_shared_policy_init(&info->policy, NULL); | |
708 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
709 | inode->i_fop = &simple_dir_operations; | |
710 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
711 | inc_nlink(inode); | |
65ed7601 | 712 | lockdep_annotate_inode_mutex_key(inode); |
7d54fa64 AV |
713 | } |
714 | return inode; | |
715 | } | |
716 | ||
b610ded7 | 717 | /* |
c8c06efa | 718 | * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never |
b610ded7 | 719 | * be taken from reclaim -- unlike regular filesystems. This needs an |
88f306b6 | 720 | * annotation because huge_pmd_share() does an allocation under hugetlb's |
c8c06efa | 721 | * i_mmap_rwsem. |
b610ded7 | 722 | */ |
c8c06efa | 723 | static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; |
b610ded7 | 724 | |
7d54fa64 AV |
725 | static struct inode *hugetlbfs_get_inode(struct super_block *sb, |
726 | struct inode *dir, | |
18df2252 | 727 | umode_t mode, dev_t dev) |
7d54fa64 AV |
728 | { |
729 | struct inode *inode; | |
9119a41e JK |
730 | struct resv_map *resv_map; |
731 | ||
732 | resv_map = resv_map_alloc(); | |
733 | if (!resv_map) | |
734 | return NULL; | |
7d54fa64 AV |
735 | |
736 | inode = new_inode(sb); | |
737 | if (inode) { | |
738 | struct hugetlbfs_inode_info *info; | |
739 | inode->i_ino = get_next_ino(); | |
740 | inode_init_owner(inode, dir, mode); | |
c8c06efa DB |
741 | lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, |
742 | &hugetlbfs_i_mmap_rwsem_key); | |
1da177e4 | 743 | inode->i_mapping->a_ops = &hugetlbfs_aops; |
1da177e4 | 744 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
9119a41e | 745 | inode->i_mapping->private_data = resv_map; |
1da177e4 | 746 | info = HUGETLBFS_I(inode); |
6bfde05b EM |
747 | /* |
748 | * The policy is initialized here even if we are creating a | |
749 | * private inode because initialization simply creates an | |
4a8c7bb5 | 750 | * an empty rb tree and calls rwlock_init(), later when we |
6bfde05b EM |
751 | * call mpol_free_shared_policy() it will just return because |
752 | * the rb tree will still be empty. | |
753 | */ | |
71fe804b | 754 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
755 | switch (mode & S_IFMT) { |
756 | default: | |
757 | init_special_inode(inode, mode, dev); | |
758 | break; | |
759 | case S_IFREG: | |
760 | inode->i_op = &hugetlbfs_inode_operations; | |
761 | inode->i_fop = &hugetlbfs_file_operations; | |
762 | break; | |
763 | case S_IFDIR: | |
764 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
765 | inode->i_fop = &simple_dir_operations; | |
766 | ||
767 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
d8c76e6f | 768 | inc_nlink(inode); |
1da177e4 LT |
769 | break; |
770 | case S_IFLNK: | |
771 | inode->i_op = &page_symlink_inode_operations; | |
21fc61c7 | 772 | inode_nohighmem(inode); |
1da177e4 LT |
773 | break; |
774 | } | |
e096d0c7 | 775 | lockdep_annotate_inode_mutex_key(inode); |
9119a41e JK |
776 | } else |
777 | kref_put(&resv_map->refs, resv_map_release); | |
778 | ||
1da177e4 LT |
779 | return inode; |
780 | } | |
781 | ||
782 | /* | |
783 | * File creation. Allocate an inode, and we're done.. | |
784 | */ | |
785 | static int hugetlbfs_mknod(struct inode *dir, | |
1a67aafb | 786 | struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 LT |
787 | { |
788 | struct inode *inode; | |
789 | int error = -ENOSPC; | |
7d54fa64 AV |
790 | |
791 | inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); | |
1da177e4 LT |
792 | if (inode) { |
793 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
794 | d_instantiate(dentry, inode); | |
795 | dget(dentry); /* Extra count - pin the dentry in core */ | |
796 | error = 0; | |
797 | } | |
798 | return error; | |
799 | } | |
800 | ||
18bb1db3 | 801 | static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
802 | { |
803 | int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); | |
804 | if (!retval) | |
d8c76e6f | 805 | inc_nlink(dir); |
1da177e4 LT |
806 | return retval; |
807 | } | |
808 | ||
ebfc3b49 | 809 | static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 LT |
810 | { |
811 | return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); | |
812 | } | |
813 | ||
814 | static int hugetlbfs_symlink(struct inode *dir, | |
815 | struct dentry *dentry, const char *symname) | |
816 | { | |
817 | struct inode *inode; | |
818 | int error = -ENOSPC; | |
1da177e4 | 819 | |
7d54fa64 | 820 | inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); |
1da177e4 LT |
821 | if (inode) { |
822 | int l = strlen(symname)+1; | |
823 | error = page_symlink(inode, symname, l); | |
824 | if (!error) { | |
825 | d_instantiate(dentry, inode); | |
826 | dget(dentry); | |
827 | } else | |
828 | iput(inode); | |
829 | } | |
830 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
831 | ||
832 | return error; | |
833 | } | |
834 | ||
835 | /* | |
6649a386 | 836 | * mark the head page dirty |
1da177e4 LT |
837 | */ |
838 | static int hugetlbfs_set_page_dirty(struct page *page) | |
839 | { | |
d85f3385 | 840 | struct page *head = compound_head(page); |
6649a386 KC |
841 | |
842 | SetPageDirty(head); | |
1da177e4 LT |
843 | return 0; |
844 | } | |
845 | ||
290408d4 | 846 | static int hugetlbfs_migrate_page(struct address_space *mapping, |
b969c4ab | 847 | struct page *newpage, struct page *page, |
a6bc32b8 | 848 | enum migrate_mode mode) |
290408d4 NH |
849 | { |
850 | int rc; | |
851 | ||
852 | rc = migrate_huge_page_move_mapping(mapping, newpage, page); | |
78bd5209 | 853 | if (rc != MIGRATEPAGE_SUCCESS) |
290408d4 NH |
854 | return rc; |
855 | migrate_page_copy(newpage, page); | |
856 | ||
78bd5209 | 857 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
858 | } |
859 | ||
726c3342 | 860 | static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 861 | { |
726c3342 | 862 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); |
2b0143b5 | 863 | struct hstate *h = hstate_inode(d_inode(dentry)); |
1da177e4 LT |
864 | |
865 | buf->f_type = HUGETLBFS_MAGIC; | |
a5516438 | 866 | buf->f_bsize = huge_page_size(h); |
1da177e4 LT |
867 | if (sbinfo) { |
868 | spin_lock(&sbinfo->stat_lock); | |
74a8a65c DG |
869 | /* If no limits set, just report 0 for max/free/used |
870 | * blocks, like simple_statfs() */ | |
90481622 DG |
871 | if (sbinfo->spool) { |
872 | long free_pages; | |
873 | ||
874 | spin_lock(&sbinfo->spool->lock); | |
875 | buf->f_blocks = sbinfo->spool->max_hpages; | |
876 | free_pages = sbinfo->spool->max_hpages | |
877 | - sbinfo->spool->used_hpages; | |
878 | buf->f_bavail = buf->f_bfree = free_pages; | |
879 | spin_unlock(&sbinfo->spool->lock); | |
74a8a65c DG |
880 | buf->f_files = sbinfo->max_inodes; |
881 | buf->f_ffree = sbinfo->free_inodes; | |
882 | } | |
1da177e4 LT |
883 | spin_unlock(&sbinfo->stat_lock); |
884 | } | |
885 | buf->f_namelen = NAME_MAX; | |
886 | return 0; | |
887 | } | |
888 | ||
889 | static void hugetlbfs_put_super(struct super_block *sb) | |
890 | { | |
891 | struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); | |
892 | ||
893 | if (sbi) { | |
894 | sb->s_fs_info = NULL; | |
90481622 DG |
895 | |
896 | if (sbi->spool) | |
897 | hugepage_put_subpool(sbi->spool); | |
898 | ||
1da177e4 LT |
899 | kfree(sbi); |
900 | } | |
901 | } | |
902 | ||
96527980 CH |
903 | static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
904 | { | |
905 | if (sbinfo->free_inodes >= 0) { | |
906 | spin_lock(&sbinfo->stat_lock); | |
907 | if (unlikely(!sbinfo->free_inodes)) { | |
908 | spin_unlock(&sbinfo->stat_lock); | |
909 | return 0; | |
910 | } | |
911 | sbinfo->free_inodes--; | |
912 | spin_unlock(&sbinfo->stat_lock); | |
913 | } | |
914 | ||
915 | return 1; | |
916 | } | |
917 | ||
918 | static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) | |
919 | { | |
920 | if (sbinfo->free_inodes >= 0) { | |
921 | spin_lock(&sbinfo->stat_lock); | |
922 | sbinfo->free_inodes++; | |
923 | spin_unlock(&sbinfo->stat_lock); | |
924 | } | |
925 | } | |
926 | ||
927 | ||
e18b890b | 928 | static struct kmem_cache *hugetlbfs_inode_cachep; |
1da177e4 LT |
929 | |
930 | static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) | |
931 | { | |
96527980 | 932 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); |
1da177e4 LT |
933 | struct hugetlbfs_inode_info *p; |
934 | ||
96527980 CH |
935 | if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) |
936 | return NULL; | |
e94b1766 | 937 | p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); |
96527980 CH |
938 | if (unlikely(!p)) { |
939 | hugetlbfs_inc_free_inodes(sbinfo); | |
1da177e4 | 940 | return NULL; |
96527980 | 941 | } |
1da177e4 LT |
942 | return &p->vfs_inode; |
943 | } | |
944 | ||
fa0d7e3d NP |
945 | static void hugetlbfs_i_callback(struct rcu_head *head) |
946 | { | |
947 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
948 | kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); |
949 | } | |
950 | ||
1da177e4 LT |
951 | static void hugetlbfs_destroy_inode(struct inode *inode) |
952 | { | |
96527980 | 953 | hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); |
1da177e4 | 954 | mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); |
fa0d7e3d | 955 | call_rcu(&inode->i_rcu, hugetlbfs_i_callback); |
1da177e4 LT |
956 | } |
957 | ||
f5e54d6e | 958 | static const struct address_space_operations hugetlbfs_aops = { |
800d15a5 NP |
959 | .write_begin = hugetlbfs_write_begin, |
960 | .write_end = hugetlbfs_write_end, | |
1da177e4 | 961 | .set_page_dirty = hugetlbfs_set_page_dirty, |
290408d4 | 962 | .migratepage = hugetlbfs_migrate_page, |
1da177e4 LT |
963 | }; |
964 | ||
96527980 | 965 | |
51cc5068 | 966 | static void init_once(void *foo) |
96527980 CH |
967 | { |
968 | struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; | |
969 | ||
a35afb83 | 970 | inode_init_once(&ei->vfs_inode); |
96527980 CH |
971 | } |
972 | ||
4b6f5d20 | 973 | const struct file_operations hugetlbfs_file_operations = { |
34d0640e | 974 | .read_iter = hugetlbfs_read_iter, |
1da177e4 | 975 | .mmap = hugetlbfs_file_mmap, |
1b061d92 | 976 | .fsync = noop_fsync, |
1da177e4 | 977 | .get_unmapped_area = hugetlb_get_unmapped_area, |
70c3547e MK |
978 | .llseek = default_llseek, |
979 | .fallocate = hugetlbfs_fallocate, | |
1da177e4 LT |
980 | }; |
981 | ||
92e1d5be | 982 | static const struct inode_operations hugetlbfs_dir_inode_operations = { |
1da177e4 LT |
983 | .create = hugetlbfs_create, |
984 | .lookup = simple_lookup, | |
985 | .link = simple_link, | |
986 | .unlink = simple_unlink, | |
987 | .symlink = hugetlbfs_symlink, | |
988 | .mkdir = hugetlbfs_mkdir, | |
989 | .rmdir = simple_rmdir, | |
990 | .mknod = hugetlbfs_mknod, | |
991 | .rename = simple_rename, | |
992 | .setattr = hugetlbfs_setattr, | |
993 | }; | |
994 | ||
92e1d5be | 995 | static const struct inode_operations hugetlbfs_inode_operations = { |
1da177e4 LT |
996 | .setattr = hugetlbfs_setattr, |
997 | }; | |
998 | ||
ee9b6d61 | 999 | static const struct super_operations hugetlbfs_ops = { |
1da177e4 LT |
1000 | .alloc_inode = hugetlbfs_alloc_inode, |
1001 | .destroy_inode = hugetlbfs_destroy_inode, | |
2bbbda30 | 1002 | .evict_inode = hugetlbfs_evict_inode, |
1da177e4 | 1003 | .statfs = hugetlbfs_statfs, |
1da177e4 | 1004 | .put_super = hugetlbfs_put_super, |
10f19a86 | 1005 | .show_options = generic_show_options, |
1da177e4 LT |
1006 | }; |
1007 | ||
7ca02d0a MK |
1008 | enum { NO_SIZE, SIZE_STD, SIZE_PERCENT }; |
1009 | ||
1010 | /* | |
1011 | * Convert size option passed from command line to number of huge pages | |
1012 | * in the pool specified by hstate. Size option could be in bytes | |
1013 | * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). | |
1014 | */ | |
1015 | static long long | |
1016 | hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, | |
1017 | int val_type) | |
1018 | { | |
1019 | if (val_type == NO_SIZE) | |
1020 | return -1; | |
1021 | ||
1022 | if (val_type == SIZE_PERCENT) { | |
1023 | size_opt <<= huge_page_shift(h); | |
1024 | size_opt *= h->max_huge_pages; | |
1025 | do_div(size_opt, 100); | |
1026 | } | |
1027 | ||
1028 | size_opt >>= huge_page_shift(h); | |
1029 | return size_opt; | |
1030 | } | |
1031 | ||
1da177e4 LT |
1032 | static int |
1033 | hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) | |
1034 | { | |
e73a75fa RD |
1035 | char *p, *rest; |
1036 | substring_t args[MAX_OPT_ARGS]; | |
1037 | int option; | |
7ca02d0a MK |
1038 | unsigned long long max_size_opt = 0, min_size_opt = 0; |
1039 | int max_val_type = NO_SIZE, min_val_type = NO_SIZE; | |
1da177e4 LT |
1040 | |
1041 | if (!options) | |
1042 | return 0; | |
1da177e4 | 1043 | |
e73a75fa RD |
1044 | while ((p = strsep(&options, ",")) != NULL) { |
1045 | int token; | |
b4c07bce LS |
1046 | if (!*p) |
1047 | continue; | |
e73a75fa RD |
1048 | |
1049 | token = match_token(p, tokens, args); | |
1050 | switch (token) { | |
1051 | case Opt_uid: | |
1052 | if (match_int(&args[0], &option)) | |
1053 | goto bad_val; | |
a0eb3a05 EB |
1054 | pconfig->uid = make_kuid(current_user_ns(), option); |
1055 | if (!uid_valid(pconfig->uid)) | |
1056 | goto bad_val; | |
e73a75fa RD |
1057 | break; |
1058 | ||
1059 | case Opt_gid: | |
1060 | if (match_int(&args[0], &option)) | |
1061 | goto bad_val; | |
a0eb3a05 EB |
1062 | pconfig->gid = make_kgid(current_user_ns(), option); |
1063 | if (!gid_valid(pconfig->gid)) | |
1064 | goto bad_val; | |
e73a75fa RD |
1065 | break; |
1066 | ||
1067 | case Opt_mode: | |
1068 | if (match_octal(&args[0], &option)) | |
1069 | goto bad_val; | |
75897d60 | 1070 | pconfig->mode = option & 01777U; |
e73a75fa RD |
1071 | break; |
1072 | ||
1073 | case Opt_size: { | |
e73a75fa RD |
1074 | /* memparse() will accept a K/M/G without a digit */ |
1075 | if (!isdigit(*args[0].from)) | |
1076 | goto bad_val; | |
7ca02d0a MK |
1077 | max_size_opt = memparse(args[0].from, &rest); |
1078 | max_val_type = SIZE_STD; | |
a137e1cc | 1079 | if (*rest == '%') |
7ca02d0a | 1080 | max_val_type = SIZE_PERCENT; |
e73a75fa RD |
1081 | break; |
1082 | } | |
1da177e4 | 1083 | |
e73a75fa RD |
1084 | case Opt_nr_inodes: |
1085 | /* memparse() will accept a K/M/G without a digit */ | |
1086 | if (!isdigit(*args[0].from)) | |
1087 | goto bad_val; | |
1088 | pconfig->nr_inodes = memparse(args[0].from, &rest); | |
1089 | break; | |
1090 | ||
a137e1cc AK |
1091 | case Opt_pagesize: { |
1092 | unsigned long ps; | |
1093 | ps = memparse(args[0].from, &rest); | |
1094 | pconfig->hstate = size_to_hstate(ps); | |
1095 | if (!pconfig->hstate) { | |
9b857d26 | 1096 | pr_err("Unsupported page size %lu MB\n", |
a137e1cc AK |
1097 | ps >> 20); |
1098 | return -EINVAL; | |
1099 | } | |
1100 | break; | |
1101 | } | |
1102 | ||
7ca02d0a MK |
1103 | case Opt_min_size: { |
1104 | /* memparse() will accept a K/M/G without a digit */ | |
1105 | if (!isdigit(*args[0].from)) | |
1106 | goto bad_val; | |
1107 | min_size_opt = memparse(args[0].from, &rest); | |
1108 | min_val_type = SIZE_STD; | |
1109 | if (*rest == '%') | |
1110 | min_val_type = SIZE_PERCENT; | |
1111 | break; | |
1112 | } | |
1113 | ||
e73a75fa | 1114 | default: |
9b857d26 | 1115 | pr_err("Bad mount option: \"%s\"\n", p); |
b4c07bce | 1116 | return -EINVAL; |
e73a75fa RD |
1117 | break; |
1118 | } | |
1da177e4 | 1119 | } |
a137e1cc | 1120 | |
7ca02d0a MK |
1121 | /* |
1122 | * Use huge page pool size (in hstate) to convert the size | |
1123 | * options to number of huge pages. If NO_SIZE, -1 is returned. | |
1124 | */ | |
1125 | pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
1126 | max_size_opt, max_val_type); | |
1127 | pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
1128 | min_size_opt, min_val_type); | |
1129 | ||
1130 | /* | |
1131 | * If max_size was specified, then min_size must be smaller | |
1132 | */ | |
1133 | if (max_val_type > NO_SIZE && | |
1134 | pconfig->min_hpages > pconfig->max_hpages) { | |
1135 | pr_err("minimum size can not be greater than maximum size\n"); | |
1136 | return -EINVAL; | |
a137e1cc AK |
1137 | } |
1138 | ||
1da177e4 | 1139 | return 0; |
e73a75fa RD |
1140 | |
1141 | bad_val: | |
9b857d26 | 1142 | pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); |
c12ddba0 | 1143 | return -EINVAL; |
1da177e4 LT |
1144 | } |
1145 | ||
1146 | static int | |
1147 | hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) | |
1148 | { | |
1da177e4 LT |
1149 | int ret; |
1150 | struct hugetlbfs_config config; | |
1151 | struct hugetlbfs_sb_info *sbinfo; | |
1152 | ||
10f19a86 MS |
1153 | save_mount_options(sb, data); |
1154 | ||
7ca02d0a | 1155 | config.max_hpages = -1; /* No limit on size by default */ |
1da177e4 | 1156 | config.nr_inodes = -1; /* No limit on number of inodes by default */ |
77c70de1 DH |
1157 | config.uid = current_fsuid(); |
1158 | config.gid = current_fsgid(); | |
1da177e4 | 1159 | config.mode = 0755; |
a137e1cc | 1160 | config.hstate = &default_hstate; |
7ca02d0a | 1161 | config.min_hpages = -1; /* No default minimum size */ |
1da177e4 | 1162 | ret = hugetlbfs_parse_options(data, &config); |
1da177e4 LT |
1163 | if (ret) |
1164 | return ret; | |
1165 | ||
1166 | sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); | |
1167 | if (!sbinfo) | |
1168 | return -ENOMEM; | |
1169 | sb->s_fs_info = sbinfo; | |
a137e1cc | 1170 | sbinfo->hstate = config.hstate; |
1da177e4 | 1171 | spin_lock_init(&sbinfo->stat_lock); |
1da177e4 LT |
1172 | sbinfo->max_inodes = config.nr_inodes; |
1173 | sbinfo->free_inodes = config.nr_inodes; | |
90481622 | 1174 | sbinfo->spool = NULL; |
7ca02d0a MK |
1175 | /* |
1176 | * Allocate and initialize subpool if maximum or minimum size is | |
1177 | * specified. Any needed reservations (for minimim size) are taken | |
1178 | * taken when the subpool is created. | |
1179 | */ | |
1180 | if (config.max_hpages != -1 || config.min_hpages != -1) { | |
1181 | sbinfo->spool = hugepage_new_subpool(config.hstate, | |
1182 | config.max_hpages, | |
1183 | config.min_hpages); | |
90481622 DG |
1184 | if (!sbinfo->spool) |
1185 | goto out_free; | |
1186 | } | |
1da177e4 | 1187 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
a137e1cc AK |
1188 | sb->s_blocksize = huge_page_size(config.hstate); |
1189 | sb->s_blocksize_bits = huge_page_shift(config.hstate); | |
1da177e4 LT |
1190 | sb->s_magic = HUGETLBFS_MAGIC; |
1191 | sb->s_op = &hugetlbfs_ops; | |
1192 | sb->s_time_gran = 1; | |
48fde701 AV |
1193 | sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); |
1194 | if (!sb->s_root) | |
1da177e4 | 1195 | goto out_free; |
1da177e4 LT |
1196 | return 0; |
1197 | out_free: | |
6e6870d4 | 1198 | kfree(sbinfo->spool); |
1da177e4 LT |
1199 | kfree(sbinfo); |
1200 | return -ENOMEM; | |
1201 | } | |
1202 | ||
3c26ff6e AV |
1203 | static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, |
1204 | int flags, const char *dev_name, void *data) | |
1da177e4 | 1205 | { |
3c26ff6e | 1206 | return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); |
1da177e4 LT |
1207 | } |
1208 | ||
1209 | static struct file_system_type hugetlbfs_fs_type = { | |
1210 | .name = "hugetlbfs", | |
3c26ff6e | 1211 | .mount = hugetlbfs_mount, |
1da177e4 LT |
1212 | .kill_sb = kill_litter_super, |
1213 | }; | |
1214 | ||
42d7395f | 1215 | static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; |
1da177e4 | 1216 | |
ef1ff6b8 | 1217 | static int can_do_hugetlb_shm(void) |
1da177e4 | 1218 | { |
a0eb3a05 EB |
1219 | kgid_t shm_group; |
1220 | shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); | |
1221 | return capable(CAP_IPC_LOCK) || in_group_p(shm_group); | |
1da177e4 LT |
1222 | } |
1223 | ||
42d7395f AK |
1224 | static int get_hstate_idx(int page_size_log) |
1225 | { | |
af73e4d9 | 1226 | struct hstate *h = hstate_sizelog(page_size_log); |
42d7395f | 1227 | |
42d7395f AK |
1228 | if (!h) |
1229 | return -1; | |
1230 | return h - hstates; | |
1231 | } | |
1232 | ||
be1d2cf5 | 1233 | static const struct dentry_operations anon_ops = { |
118b2302 | 1234 | .d_dname = simple_dname |
0df4d6e5 AV |
1235 | }; |
1236 | ||
af73e4d9 NH |
1237 | /* |
1238 | * Note that size should be aligned to proper hugepage size in caller side, | |
1239 | * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. | |
1240 | */ | |
1241 | struct file *hugetlb_file_setup(const char *name, size_t size, | |
1242 | vm_flags_t acctflag, struct user_struct **user, | |
42d7395f | 1243 | int creat_flags, int page_size_log) |
1da177e4 | 1244 | { |
39b65252 | 1245 | struct file *file = ERR_PTR(-ENOMEM); |
1da177e4 | 1246 | struct inode *inode; |
2c48b9c4 | 1247 | struct path path; |
0df4d6e5 | 1248 | struct super_block *sb; |
1da177e4 | 1249 | struct qstr quick_string; |
42d7395f AK |
1250 | int hstate_idx; |
1251 | ||
1252 | hstate_idx = get_hstate_idx(page_size_log); | |
1253 | if (hstate_idx < 0) | |
1254 | return ERR_PTR(-ENODEV); | |
1da177e4 | 1255 | |
353d5c30 | 1256 | *user = NULL; |
42d7395f | 1257 | if (!hugetlbfs_vfsmount[hstate_idx]) |
5bc98594 AM |
1258 | return ERR_PTR(-ENOENT); |
1259 | ||
ef1ff6b8 | 1260 | if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { |
353d5c30 HD |
1261 | *user = current_user(); |
1262 | if (user_shm_lock(size, *user)) { | |
21a3c273 | 1263 | task_lock(current); |
9b857d26 | 1264 | pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", |
21a3c273 DR |
1265 | current->comm, current->pid); |
1266 | task_unlock(current); | |
353d5c30 HD |
1267 | } else { |
1268 | *user = NULL; | |
2584e517 | 1269 | return ERR_PTR(-EPERM); |
353d5c30 | 1270 | } |
2584e517 | 1271 | } |
1da177e4 | 1272 | |
0df4d6e5 | 1273 | sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; |
9d66586f | 1274 | quick_string.name = name; |
1da177e4 LT |
1275 | quick_string.len = strlen(quick_string.name); |
1276 | quick_string.hash = 0; | |
0df4d6e5 | 1277 | path.dentry = d_alloc_pseudo(sb, &quick_string); |
2c48b9c4 | 1278 | if (!path.dentry) |
1da177e4 LT |
1279 | goto out_shm_unlock; |
1280 | ||
0df4d6e5 | 1281 | d_set_d_op(path.dentry, &anon_ops); |
42d7395f | 1282 | path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); |
39b65252 | 1283 | file = ERR_PTR(-ENOSPC); |
0df4d6e5 | 1284 | inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); |
1da177e4 | 1285 | if (!inode) |
ce8d2cdf | 1286 | goto out_dentry; |
e1832f29 SS |
1287 | if (creat_flags == HUGETLB_SHMFS_INODE) |
1288 | inode->i_flags |= S_PRIVATE; | |
1da177e4 | 1289 | |
39b65252 | 1290 | file = ERR_PTR(-ENOMEM); |
af73e4d9 NH |
1291 | if (hugetlb_reserve_pages(inode, 0, |
1292 | size >> huge_page_shift(hstate_inode(inode)), NULL, | |
1293 | acctflag)) | |
b45b5bd6 DG |
1294 | goto out_inode; |
1295 | ||
2c48b9c4 | 1296 | d_instantiate(path.dentry, inode); |
1da177e4 | 1297 | inode->i_size = size; |
6d6b77f1 | 1298 | clear_nlink(inode); |
ce8d2cdf | 1299 | |
2c48b9c4 | 1300 | file = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
ce8d2cdf | 1301 | &hugetlbfs_file_operations); |
39b65252 | 1302 | if (IS_ERR(file)) |
b4d232e6 | 1303 | goto out_dentry; /* inode is already attached */ |
ce8d2cdf | 1304 | |
1da177e4 LT |
1305 | return file; |
1306 | ||
b45b5bd6 DG |
1307 | out_inode: |
1308 | iput(inode); | |
1da177e4 | 1309 | out_dentry: |
2c48b9c4 | 1310 | path_put(&path); |
1da177e4 | 1311 | out_shm_unlock: |
353d5c30 HD |
1312 | if (*user) { |
1313 | user_shm_unlock(size, *user); | |
1314 | *user = NULL; | |
1315 | } | |
39b65252 | 1316 | return file; |
1da177e4 LT |
1317 | } |
1318 | ||
1319 | static int __init init_hugetlbfs_fs(void) | |
1320 | { | |
42d7395f | 1321 | struct hstate *h; |
1da177e4 | 1322 | int error; |
42d7395f | 1323 | int i; |
1da177e4 | 1324 | |
457c1b27 | 1325 | if (!hugepages_supported()) { |
9b857d26 | 1326 | pr_info("disabling because there are no supported hugepage sizes\n"); |
457c1b27 NA |
1327 | return -ENOTSUPP; |
1328 | } | |
1329 | ||
d1d5e05f | 1330 | error = -ENOMEM; |
1da177e4 LT |
1331 | hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", |
1332 | sizeof(struct hugetlbfs_inode_info), | |
5d097056 | 1333 | 0, SLAB_ACCOUNT, init_once); |
1da177e4 | 1334 | if (hugetlbfs_inode_cachep == NULL) |
e0bf68dd | 1335 | goto out2; |
1da177e4 LT |
1336 | |
1337 | error = register_filesystem(&hugetlbfs_fs_type); | |
1338 | if (error) | |
1339 | goto out; | |
1340 | ||
42d7395f AK |
1341 | i = 0; |
1342 | for_each_hstate(h) { | |
1343 | char buf[50]; | |
1344 | unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); | |
1da177e4 | 1345 | |
42d7395f AK |
1346 | snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); |
1347 | hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, | |
1348 | buf); | |
1da177e4 | 1349 | |
42d7395f | 1350 | if (IS_ERR(hugetlbfs_vfsmount[i])) { |
9b857d26 | 1351 | pr_err("Cannot mount internal hugetlbfs for " |
42d7395f AK |
1352 | "page size %uK", ps_kb); |
1353 | error = PTR_ERR(hugetlbfs_vfsmount[i]); | |
1354 | hugetlbfs_vfsmount[i] = NULL; | |
1355 | } | |
1356 | i++; | |
1357 | } | |
1358 | /* Non default hstates are optional */ | |
1359 | if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) | |
1360 | return 0; | |
1da177e4 LT |
1361 | |
1362 | out: | |
d1d5e05f | 1363 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
e0bf68dd | 1364 | out2: |
1da177e4 LT |
1365 | return error; |
1366 | } | |
3e89e1c5 | 1367 | fs_initcall(init_hugetlbfs_fs) |