blob: 8349a899912e5c47ca26c66df1be66c684bdce4d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * hugetlbpage-backed filesystem. Based on ramfs.
3 *
4 * William Irwin, 2002
5 *
6 * Copyright (C) 2002 Linus Torvalds.
7 */
8
9#include <linux/module.h>
10#include <linux/thread_info.h>
11#include <asm/current.h>
12#include <linux/sched.h> /* remove ASAP */
13#include <linux/fs.h>
14#include <linux/mount.h>
15#include <linux/file.h>
Randy Dunlape73a75f2007-07-15 23:40:52 -070016#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/writeback.h>
18#include <linux/pagemap.h>
19#include <linux/highmem.h>
20#include <linux/init.h>
21#include <linux/string.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080022#include <linux/capability.h>
Randy Dunlape73a75f2007-07-15 23:40:52 -070023#include <linux/ctype.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/backing-dev.h>
25#include <linux/hugetlb.h>
26#include <linux/pagevec.h>
Randy Dunlape73a75f2007-07-15 23:40:52 -070027#include <linux/parser.h>
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -070028#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/slab.h>
30#include <linux/dnotify.h>
31#include <linux/statfs.h>
32#include <linux/security.h>
Nick Black1fd7317d2009-09-22 16:43:33 -070033#include <linux/magic.h>
Naoya Horiguchi290408d2010-09-08 10:19:35 +090034#include <linux/migrate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36#include <asm/uaccess.h>
37
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -080038static const struct super_operations hugetlbfs_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070039static const struct address_space_operations hugetlbfs_aops;
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080040const struct file_operations hugetlbfs_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080041static const struct inode_operations hugetlbfs_dir_inode_operations;
42static const struct inode_operations hugetlbfs_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
David Gibsona1d776e2012-03-21 16:34:12 -070044struct hugetlbfs_config {
45 uid_t uid;
46 gid_t gid;
47 umode_t mode;
48 long nr_blocks;
49 long nr_inodes;
50 struct hstate *hstate;
51};
52
53struct hugetlbfs_inode_info {
54 struct shared_policy policy;
55 struct inode vfs_inode;
56};
57
58static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
59{
60 return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
61}
62
Linus Torvalds1da177e2005-04-16 15:20:36 -070063static struct backing_dev_info hugetlbfs_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +020064 .name = "hugetlbfs",
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 .ra_pages = 0, /* No readahead */
Miklos Szeredie4ad08f2008-04-30 00:54:37 -070066 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Linus Torvalds1da177e2005-04-16 15:20:36 -070067};
68
69int sysctl_hugetlb_shm_group;
70
Randy Dunlape73a75f2007-07-15 23:40:52 -070071enum {
72 Opt_size, Opt_nr_inodes,
73 Opt_mode, Opt_uid, Opt_gid,
Andi Kleena137e1c2008-07-23 21:27:43 -070074 Opt_pagesize,
Randy Dunlape73a75f2007-07-15 23:40:52 -070075 Opt_err,
76};
77
Steven Whitehousea447c092008-10-13 10:46:57 +010078static const match_table_t tokens = {
Randy Dunlape73a75f2007-07-15 23:40:52 -070079 {Opt_size, "size=%s"},
80 {Opt_nr_inodes, "nr_inodes=%s"},
81 {Opt_mode, "mode=%o"},
82 {Opt_uid, "uid=%u"},
83 {Opt_gid, "gid=%u"},
Andi Kleena137e1c2008-07-23 21:27:43 -070084 {Opt_pagesize, "pagesize=%s"},
Randy Dunlape73a75f2007-07-15 23:40:52 -070085 {Opt_err, NULL},
86};
87
Adam Litke2e9b3672005-10-29 18:16:47 -070088static void huge_pagevec_release(struct pagevec *pvec)
89{
90 int i;
91
92 for (i = 0; i < pagevec_count(pvec); ++i)
93 put_page(pvec->pages[i]);
94
95 pagevec_reinit(pvec);
96}
97
Linus Torvalds1da177e2005-04-16 15:20:36 -070098static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
99{
Josef Sipekb39424e2006-12-08 02:37:07 -0800100 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 loff_t len, vma_len;
102 int ret;
Andi Kleena5516432008-07-23 21:27:41 -0700103 struct hstate *h = hstate_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
Hugh Dickins68589bc2006-11-14 02:03:32 -0800105 /*
David Gibsondec4ad82007-08-30 23:56:40 -0700106 * vma address alignment (but not the pgoff alignment) has
107 * already been checked by prepare_hugepage_range. If you add
108 * any error returns here, do so after setting VM_HUGETLB, so
109 * is_vm_hugetlb_page tests below unmap_region go the right
110 * way when do_mmap_pgoff unwinds (may be important on powerpc
111 * and ia64).
Hugh Dickins68589bc2006-11-14 02:03:32 -0800112 */
113 vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
114 vma->vm_ops = &hugetlb_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Becky Bruce2b37c352011-07-25 17:11:49 -0700116 if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
David Gibsondec4ad82007-08-30 23:56:40 -0700117 return -EINVAL;
118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
120
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800121 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 file_accessed(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124 ret = -ENOMEM;
125 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Mel Gormana1e78772008-07-23 21:27:23 -0700127 if (hugetlb_reserve_pages(inode,
Andi Kleena5516432008-07-23 21:27:41 -0700128 vma->vm_pgoff >> huge_page_order(h),
Mel Gorman5a6fe122009-02-10 14:02:27 +0000129 len >> huge_page_shift(h), vma,
130 vma->vm_flags))
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700131 goto out;
David Gibsonb45b5bd2006-03-22 00:08:55 -0800132
Adam Litke4c887262005-10-29 18:16:46 -0700133 ret = 0;
134 hugetlb_prefault_arch_hook(vma->vm_mm);
Zhang, Yanminb6174df2006-07-10 04:44:49 -0700135 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 inode->i_size = len;
137out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800138 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
140 return ret;
141}
142
143/*
Hugh Dickins508034a2005-10-29 18:16:30 -0700144 * Called under down_write(mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 */
146
Adrian Bunkd2ba27e82007-05-06 14:49:00 -0700147#ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148static unsigned long
149hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
150 unsigned long len, unsigned long pgoff, unsigned long flags)
151{
152 struct mm_struct *mm = current->mm;
153 struct vm_area_struct *vma;
154 unsigned long start_addr;
Andi Kleena5516432008-07-23 21:27:41 -0700155 struct hstate *h = hstate_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Andi Kleena5516432008-07-23 21:27:41 -0700157 if (len & ~huge_page_mask(h))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 return -EINVAL;
159 if (len > TASK_SIZE)
160 return -ENOMEM;
161
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -0700162 if (flags & MAP_FIXED) {
Andi Kleena5516432008-07-23 21:27:41 -0700163 if (prepare_hugepage_range(file, addr, len))
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -0700164 return -EINVAL;
165 return addr;
166 }
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 if (addr) {
Andi Kleena5516432008-07-23 21:27:41 -0700169 addr = ALIGN(addr, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 vma = find_vma(mm, addr);
171 if (TASK_SIZE - len >= addr &&
172 (!vma || addr + len <= vma->vm_start))
173 return addr;
174 }
175
Xiao Guangrong4bfc1302012-03-21 16:33:54 -0700176 if (len > mm->cached_hole_size)
177 start_addr = mm->free_area_cache;
178 else {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -0700179 start_addr = TASK_UNMAPPED_BASE;
Xiao Guangrong4bfc1302012-03-21 16:33:54 -0700180 mm->cached_hole_size = 0;
181 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -0700182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183full_search:
Andi Kleena5516432008-07-23 21:27:41 -0700184 addr = ALIGN(start_addr, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
186 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
187 /* At this point: (!vma || addr < vma->vm_end). */
188 if (TASK_SIZE - len < addr) {
189 /*
190 * Start a new search - just in case we missed
191 * some holes.
192 */
193 if (start_addr != TASK_UNMAPPED_BASE) {
194 start_addr = TASK_UNMAPPED_BASE;
Xiao Guangrong4bfc1302012-03-21 16:33:54 -0700195 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 goto full_search;
197 }
198 return -ENOMEM;
199 }
200
Xiao Guangrong4bfc1302012-03-21 16:33:54 -0700201 if (!vma || addr + len <= vma->vm_start) {
202 mm->free_area_cache = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 return addr;
Xiao Guangrong4bfc1302012-03-21 16:33:54 -0700204 }
205 if (addr + mm->cached_hole_size < vma->vm_start)
206 mm->cached_hole_size = vma->vm_start - addr;
Andi Kleena5516432008-07-23 21:27:41 -0700207 addr = ALIGN(vma->vm_end, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 }
209}
210#endif
211
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700212static int
213hugetlbfs_read_actor(struct page *page, unsigned long offset,
214 char __user *buf, unsigned long count,
215 unsigned long size)
216{
217 char *kaddr;
218 unsigned long left, copied = 0;
219 int i, chunksize;
220
221 if (size > count)
222 size = count;
223
224 /* Find which 4k chunk and offset with in that chunk */
225 i = offset >> PAGE_CACHE_SHIFT;
226 offset = offset & ~PAGE_CACHE_MASK;
227
228 while (size) {
229 chunksize = PAGE_CACHE_SIZE;
230 if (offset)
231 chunksize -= offset;
232 if (chunksize > size)
233 chunksize = size;
234 kaddr = kmap(&page[i]);
235 left = __copy_to_user(buf, kaddr + offset, chunksize);
236 kunmap(&page[i]);
237 if (left) {
238 copied += (chunksize - left);
239 break;
240 }
241 offset = 0;
242 size -= chunksize;
243 buf += chunksize;
244 copied += chunksize;
245 i++;
246 }
247 return copied ? copied : -EFAULT;
248}
249
250/*
251 * Support for read() - Find the page attached to f_mapping and copy out the
252 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
253 * since it has PAGE_CACHE_SIZE assumptions.
254 */
255static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
256 size_t len, loff_t *ppos)
257{
Andi Kleena5516432008-07-23 21:27:41 -0700258 struct hstate *h = hstate_file(filp);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700259 struct address_space *mapping = filp->f_mapping;
260 struct inode *inode = mapping->host;
Andi Kleena5516432008-07-23 21:27:41 -0700261 unsigned long index = *ppos >> huge_page_shift(h);
262 unsigned long offset = *ppos & ~huge_page_mask(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700263 unsigned long end_index;
264 loff_t isize;
265 ssize_t retval = 0;
266
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700267 /* validate length */
268 if (len == 0)
269 goto out;
270
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700271 for (;;) {
272 struct page *page;
Andi Kleena5516432008-07-23 21:27:41 -0700273 unsigned long nr, ret;
Roel Kluin91bf189c2009-01-06 14:40:14 -0800274 int ra;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700275
276 /* nr is the maximum number of bytes to copy from this page */
Andi Kleena5516432008-07-23 21:27:41 -0700277 nr = huge_page_size(h);
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700278 isize = i_size_read(inode);
279 if (!isize)
280 goto out;
281 end_index = (isize - 1) >> huge_page_shift(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700282 if (index >= end_index) {
283 if (index > end_index)
284 goto out;
Andi Kleena5516432008-07-23 21:27:41 -0700285 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700286 if (nr <= offset)
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700287 goto out;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700288 }
289 nr = nr - offset;
290
291 /* Find the page */
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700292 page = find_lock_page(mapping, index);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700293 if (unlikely(page == NULL)) {
294 /*
295 * We have a HOLE, zero out the user-buffer for the
296 * length of the hole or request.
297 */
298 ret = len < nr ? len : nr;
299 if (clear_user(buf, ret))
Roel Kluin91bf189c2009-01-06 14:40:14 -0800300 ra = -EFAULT;
301 else
302 ra = 0;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700303 } else {
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700304 unlock_page(page);
305
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700306 /*
307 * We have the page, copy it to user space buffer.
308 */
Roel Kluin91bf189c2009-01-06 14:40:14 -0800309 ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
310 ret = ra;
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700311 page_cache_release(page);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700312 }
Roel Kluin91bf189c2009-01-06 14:40:14 -0800313 if (ra < 0) {
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700314 if (retval == 0)
Roel Kluin91bf189c2009-01-06 14:40:14 -0800315 retval = ra;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700316 goto out;
317 }
318
319 offset += ret;
320 retval += ret;
321 len -= ret;
Andi Kleena5516432008-07-23 21:27:41 -0700322 index += offset >> huge_page_shift(h);
323 offset &= ~huge_page_mask(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700324
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700325 /* short read or no more work */
326 if ((ret != nr) || (len == 0))
327 break;
328 }
329out:
Andi Kleena5516432008-07-23 21:27:41 -0700330 *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700331 return retval;
332}
333
Nick Piggin800d15a2007-10-16 01:25:03 -0700334static int hugetlbfs_write_begin(struct file *file,
335 struct address_space *mapping,
336 loff_t pos, unsigned len, unsigned flags,
337 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338{
339 return -EINVAL;
340}
341
Nick Piggin800d15a2007-10-16 01:25:03 -0700342static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
343 loff_t pos, unsigned len, unsigned copied,
344 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
Nick Piggin800d15a2007-10-16 01:25:03 -0700346 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 return -EINVAL;
348}
349
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350static void truncate_huge_page(struct page *page)
351{
Linus Torvaldsfba25912006-12-20 13:46:42 -0800352 cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 ClearPageUptodate(page);
Minchan Kimbd65cb82011-03-22 16:30:54 -0700354 delete_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355}
356
David Gibsonb45b5bd2006-03-22 00:08:55 -0800357static void truncate_hugepages(struct inode *inode, loff_t lstart)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358{
Andi Kleena5516432008-07-23 21:27:41 -0700359 struct hstate *h = hstate_inode(inode);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800360 struct address_space *mapping = &inode->i_data;
Andi Kleena5516432008-07-23 21:27:41 -0700361 const pgoff_t start = lstart >> huge_page_shift(h);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 struct pagevec pvec;
363 pgoff_t next;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700364 int i, freed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
366 pagevec_init(&pvec, 0);
367 next = start;
368 while (1) {
369 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
370 if (next == start)
371 break;
372 next = start;
373 continue;
374 }
375
376 for (i = 0; i < pagevec_count(&pvec); ++i) {
377 struct page *page = pvec.pages[i];
378
379 lock_page(page);
380 if (page->index > next)
381 next = page->index;
382 ++next;
383 truncate_huge_page(page);
384 unlock_page(page);
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700385 freed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 }
387 huge_pagevec_release(&pvec);
388 }
389 BUG_ON(!lstart && mapping->nrpages);
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700390 hugetlb_unreserve_pages(inode, start, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
Al Viro2bbbda32010-06-04 19:52:12 -0400393static void hugetlbfs_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
David Gibsonb45b5bd2006-03-22 00:08:55 -0800395 truncate_hugepages(inode, 0);
Jan Karadbd57682012-05-03 14:48:02 +0200396 clear_inode(inode);
Christoph Hellwig149f4212005-10-29 18:16:43 -0700397}
398
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399static inline void
Hugh Dickins856fc292006-10-28 10:38:43 -0700400hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
402 struct vm_area_struct *vma;
403 struct prio_tree_iter iter;
404
Hugh Dickins856fc292006-10-28 10:38:43 -0700405 vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 unsigned long v_offset;
407
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 /*
Hugh Dickins856fc292006-10-28 10:38:43 -0700409 * Can the expression below overflow on 32-bit arches?
410 * No, because the prio_tree returns us only those vmas
411 * which overlap the truncated area starting at pgoff,
412 * and no vma on a 32-bit arch can span beyond the 4GB.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 */
Hugh Dickins856fc292006-10-28 10:38:43 -0700414 if (vma->vm_pgoff < pgoff)
415 v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
416 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 v_offset = 0;
418
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -0700419 unmap_hugepage_range(vma, vma->vm_start + v_offset,
420 vma->vm_end, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
422}
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
425{
Hugh Dickins856fc292006-10-28 10:38:43 -0700426 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 struct address_space *mapping = inode->i_mapping;
Andi Kleena5516432008-07-23 21:27:41 -0700428 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
Andi Kleena5516432008-07-23 21:27:41 -0700430 BUG_ON(offset & ~huge_page_mask(h));
Hugh Dickins856fc292006-10-28 10:38:43 -0700431 pgoff = offset >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Ken Chen7aa91e12007-10-16 01:26:21 -0700433 i_size_write(inode, offset);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700434 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 if (!prio_tree_empty(&mapping->i_mmap))
436 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700437 mutex_unlock(&mapping->i_mmap_mutex);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800438 truncate_hugepages(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 return 0;
440}
441
442static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
443{
444 struct inode *inode = dentry->d_inode;
Andi Kleena5516432008-07-23 21:27:41 -0700445 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 int error;
447 unsigned int ia_valid = attr->ia_valid;
448
449 BUG_ON(!inode);
450
451 error = inode_change_ok(inode, attr);
452 if (error)
Christoph Hellwig10257742010-06-04 11:30:02 +0200453 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455 if (ia_valid & ATTR_SIZE) {
456 error = -EINVAL;
Christoph Hellwig10257742010-06-04 11:30:02 +0200457 if (attr->ia_size & ~huge_page_mask(h))
458 return -EINVAL;
459 error = hugetlb_vmtruncate(inode, attr->ia_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 if (error)
Christoph Hellwig10257742010-06-04 11:30:02 +0200461 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
Christoph Hellwig10257742010-06-04 11:30:02 +0200463
464 setattr_copy(inode, attr);
465 mark_inode_dirty(inode);
466 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
Al Viro7d54fa62011-07-24 20:20:48 -0400469static struct inode *hugetlbfs_get_root(struct super_block *sb,
470 struct hugetlbfs_config *config)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
472 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474 inode = new_inode(sb);
475 if (inode) {
476 struct hugetlbfs_inode_info *info;
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400477 inode->i_ino = get_next_ino();
Al Viro7d54fa62011-07-24 20:20:48 -0400478 inode->i_mode = S_IFDIR | config->mode;
479 inode->i_uid = config->uid;
480 inode->i_gid = config->gid;
481 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
482 info = HUGETLBFS_I(inode);
483 mpol_shared_policy_init(&info->policy, NULL);
484 inode->i_op = &hugetlbfs_dir_inode_operations;
485 inode->i_fop = &simple_dir_operations;
486 /* directory inodes start off with i_nlink == 2 (for "." entry) */
487 inc_nlink(inode);
Aneesh Kumar K.V65ed7602012-04-25 16:01:50 -0700488 lockdep_annotate_inode_mutex_key(inode);
Al Viro7d54fa62011-07-24 20:20:48 -0400489 }
490 return inode;
491}
492
493static struct inode *hugetlbfs_get_inode(struct super_block *sb,
494 struct inode *dir,
Al Viro18df2252011-07-24 23:17:40 -0400495 umode_t mode, dev_t dev)
Al Viro7d54fa62011-07-24 20:20:48 -0400496{
497 struct inode *inode;
498
499 inode = new_inode(sb);
500 if (inode) {
501 struct hugetlbfs_inode_info *info;
502 inode->i_ino = get_next_ino();
503 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 inode->i_mapping->a_ops = &hugetlbfs_aops;
505 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
506 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700507 INIT_LIST_HEAD(&inode->i_mapping->private_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 info = HUGETLBFS_I(inode);
Eric B Munson6bfde052009-09-21 17:03:43 -0700509 /*
510 * The policy is initialized here even if we are creating a
511 * private inode because initialization simply creates an
512 * an empty rb tree and calls spin_lock_init(), later when we
513 * call mpol_free_shared_policy() it will just return because
514 * the rb tree will still be empty.
515 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700516 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 switch (mode & S_IFMT) {
518 default:
519 init_special_inode(inode, mode, dev);
520 break;
521 case S_IFREG:
522 inode->i_op = &hugetlbfs_inode_operations;
523 inode->i_fop = &hugetlbfs_file_operations;
524 break;
525 case S_IFDIR:
526 inode->i_op = &hugetlbfs_dir_inode_operations;
527 inode->i_fop = &simple_dir_operations;
528
529 /* directory inodes start off with i_nlink == 2 (for "." entry) */
Dave Hansend8c76e62006-09-30 23:29:04 -0700530 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 break;
532 case S_IFLNK:
533 inode->i_op = &page_symlink_inode_operations;
534 break;
535 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400536 lockdep_annotate_inode_mutex_key(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 }
538 return inode;
539}
540
541/*
542 * File creation. Allocate an inode, and we're done..
543 */
544static int hugetlbfs_mknod(struct inode *dir,
Al Viro1a67aaf2011-07-26 01:52:52 -0400545 struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
547 struct inode *inode;
548 int error = -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Al Viro7d54fa62011-07-24 20:20:48 -0400550 inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 if (inode) {
552 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
553 d_instantiate(dentry, inode);
554 dget(dentry); /* Extra count - pin the dentry in core */
555 error = 0;
556 }
557 return error;
558}
559
Al Viro18bb1db2011-07-26 01:41:39 -0400560static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561{
562 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
563 if (!retval)
Dave Hansend8c76e62006-09-30 23:29:04 -0700564 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 return retval;
566}
567
Al Viroebfc3b42012-06-10 18:05:36 -0400568static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
570 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
571}
572
573static int hugetlbfs_symlink(struct inode *dir,
574 struct dentry *dentry, const char *symname)
575{
576 struct inode *inode;
577 int error = -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Al Viro7d54fa62011-07-24 20:20:48 -0400579 inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 if (inode) {
581 int l = strlen(symname)+1;
582 error = page_symlink(inode, symname, l);
583 if (!error) {
584 d_instantiate(dentry, inode);
585 dget(dentry);
586 } else
587 iput(inode);
588 }
589 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
590
591 return error;
592}
593
594/*
Ken Chen6649a382007-02-08 14:20:27 -0800595 * mark the head page dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 */
597static int hugetlbfs_set_page_dirty(struct page *page)
598{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700599 struct page *head = compound_head(page);
Ken Chen6649a382007-02-08 14:20:27 -0800600
601 SetPageDirty(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 return 0;
603}
604
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900605static int hugetlbfs_migrate_page(struct address_space *mapping,
Mel Gormanb969c4a2012-01-12 17:19:34 -0800606 struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800607 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900608{
609 int rc;
610
611 rc = migrate_huge_page_move_mapping(mapping, newpage, page);
612 if (rc)
613 return rc;
614 migrate_page_copy(newpage, page);
615
616 return 0;
617}
618
David Howells726c3342006-06-23 02:02:58 -0700619static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620{
David Howells726c3342006-06-23 02:02:58 -0700621 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
Andi Kleena5516432008-07-23 21:27:41 -0700622 struct hstate *h = hstate_inode(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624 buf->f_type = HUGETLBFS_MAGIC;
Andi Kleena5516432008-07-23 21:27:41 -0700625 buf->f_bsize = huge_page_size(h);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 if (sbinfo) {
627 spin_lock(&sbinfo->stat_lock);
David Gibson74a8a652005-11-21 21:32:24 -0800628 /* If no limits set, just report 0 for max/free/used
629 * blocks, like simple_statfs() */
David Gibson90481622012-03-21 16:34:12 -0700630 if (sbinfo->spool) {
631 long free_pages;
632
633 spin_lock(&sbinfo->spool->lock);
634 buf->f_blocks = sbinfo->spool->max_hpages;
635 free_pages = sbinfo->spool->max_hpages
636 - sbinfo->spool->used_hpages;
637 buf->f_bavail = buf->f_bfree = free_pages;
638 spin_unlock(&sbinfo->spool->lock);
David Gibson74a8a652005-11-21 21:32:24 -0800639 buf->f_files = sbinfo->max_inodes;
640 buf->f_ffree = sbinfo->free_inodes;
641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 spin_unlock(&sbinfo->stat_lock);
643 }
644 buf->f_namelen = NAME_MAX;
645 return 0;
646}
647
648static void hugetlbfs_put_super(struct super_block *sb)
649{
650 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
651
652 if (sbi) {
653 sb->s_fs_info = NULL;
David Gibson90481622012-03-21 16:34:12 -0700654
655 if (sbi->spool)
656 hugepage_put_subpool(sbi->spool);
657
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 kfree(sbi);
659 }
660}
661
Christoph Hellwig96527982005-10-29 18:16:42 -0700662static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
663{
664 if (sbinfo->free_inodes >= 0) {
665 spin_lock(&sbinfo->stat_lock);
666 if (unlikely(!sbinfo->free_inodes)) {
667 spin_unlock(&sbinfo->stat_lock);
668 return 0;
669 }
670 sbinfo->free_inodes--;
671 spin_unlock(&sbinfo->stat_lock);
672 }
673
674 return 1;
675}
676
677static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
678{
679 if (sbinfo->free_inodes >= 0) {
680 spin_lock(&sbinfo->stat_lock);
681 sbinfo->free_inodes++;
682 spin_unlock(&sbinfo->stat_lock);
683 }
684}
685
686
Christoph Lametere18b8902006-12-06 20:33:20 -0800687static struct kmem_cache *hugetlbfs_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
689static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
690{
Christoph Hellwig96527982005-10-29 18:16:42 -0700691 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 struct hugetlbfs_inode_info *p;
693
Christoph Hellwig96527982005-10-29 18:16:42 -0700694 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 return NULL;
Christoph Lametere94b1762006-12-06 20:33:17 -0800696 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
Christoph Hellwig96527982005-10-29 18:16:42 -0700697 if (unlikely(!p)) {
698 hugetlbfs_inc_free_inodes(sbinfo);
699 return NULL;
700 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 return &p->vfs_inode;
702}
703
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100704static void hugetlbfs_i_callback(struct rcu_head *head)
705{
706 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100707 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
708}
709
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710static void hugetlbfs_destroy_inode(struct inode *inode)
711{
Christoph Hellwig96527982005-10-29 18:16:42 -0700712 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100714 call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715}
716
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700717static const struct address_space_operations hugetlbfs_aops = {
Nick Piggin800d15a2007-10-16 01:25:03 -0700718 .write_begin = hugetlbfs_write_begin,
719 .write_end = hugetlbfs_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 .set_page_dirty = hugetlbfs_set_page_dirty,
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900721 .migratepage = hugetlbfs_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722};
723
Christoph Hellwig96527982005-10-29 18:16:42 -0700724
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700725static void init_once(void *foo)
Christoph Hellwig96527982005-10-29 18:16:42 -0700726{
727 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
728
Christoph Lametera35afb82007-05-16 22:10:57 -0700729 inode_init_once(&ei->vfs_inode);
Christoph Hellwig96527982005-10-29 18:16:42 -0700730}
731
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800732const struct file_operations hugetlbfs_file_operations = {
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700733 .read = hugetlbfs_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 .mmap = hugetlbfs_file_mmap,
Christoph Hellwig1b061d92010-05-26 17:53:41 +0200735 .fsync = noop_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 .get_unmapped_area = hugetlb_get_unmapped_area,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200737 .llseek = default_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738};
739
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800740static const struct inode_operations hugetlbfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 .create = hugetlbfs_create,
742 .lookup = simple_lookup,
743 .link = simple_link,
744 .unlink = simple_unlink,
745 .symlink = hugetlbfs_symlink,
746 .mkdir = hugetlbfs_mkdir,
747 .rmdir = simple_rmdir,
748 .mknod = hugetlbfs_mknod,
749 .rename = simple_rename,
750 .setattr = hugetlbfs_setattr,
751};
752
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800753static const struct inode_operations hugetlbfs_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 .setattr = hugetlbfs_setattr,
755};
756
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800757static const struct super_operations hugetlbfs_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 .alloc_inode = hugetlbfs_alloc_inode,
759 .destroy_inode = hugetlbfs_destroy_inode,
Al Viro2bbbda32010-06-04 19:52:12 -0400760 .evict_inode = hugetlbfs_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 .statfs = hugetlbfs_statfs,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 .put_super = hugetlbfs_put_super,
Miklos Szeredi10f19a82008-02-08 04:21:45 -0800763 .show_options = generic_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764};
765
766static int
767hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
768{
Randy Dunlape73a75f2007-07-15 23:40:52 -0700769 char *p, *rest;
770 substring_t args[MAX_OPT_ARGS];
771 int option;
Andi Kleena137e1c2008-07-23 21:27:43 -0700772 unsigned long long size = 0;
773 enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
775 if (!options)
776 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Randy Dunlape73a75f2007-07-15 23:40:52 -0700778 while ((p = strsep(&options, ",")) != NULL) {
779 int token;
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -0700780 if (!*p)
781 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
Randy Dunlape73a75f2007-07-15 23:40:52 -0700783 token = match_token(p, tokens, args);
784 switch (token) {
785 case Opt_uid:
786 if (match_int(&args[0], &option))
787 goto bad_val;
788 pconfig->uid = option;
789 break;
790
791 case Opt_gid:
792 if (match_int(&args[0], &option))
793 goto bad_val;
794 pconfig->gid = option;
795 break;
796
797 case Opt_mode:
798 if (match_octal(&args[0], &option))
799 goto bad_val;
Ken Chen75897d62008-02-04 22:28:36 -0800800 pconfig->mode = option & 01777U;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700801 break;
802
803 case Opt_size: {
Randy Dunlape73a75f2007-07-15 23:40:52 -0700804 /* memparse() will accept a K/M/G without a digit */
805 if (!isdigit(*args[0].from))
806 goto bad_val;
807 size = memparse(args[0].from, &rest);
Andi Kleena137e1c2008-07-23 21:27:43 -0700808 setsize = SIZE_STD;
809 if (*rest == '%')
810 setsize = SIZE_PERCENT;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700811 break;
812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Randy Dunlape73a75f2007-07-15 23:40:52 -0700814 case Opt_nr_inodes:
815 /* memparse() will accept a K/M/G without a digit */
816 if (!isdigit(*args[0].from))
817 goto bad_val;
818 pconfig->nr_inodes = memparse(args[0].from, &rest);
819 break;
820
Andi Kleena137e1c2008-07-23 21:27:43 -0700821 case Opt_pagesize: {
822 unsigned long ps;
823 ps = memparse(args[0].from, &rest);
824 pconfig->hstate = size_to_hstate(ps);
825 if (!pconfig->hstate) {
826 printk(KERN_ERR
827 "hugetlbfs: Unsupported page size %lu MB\n",
828 ps >> 20);
829 return -EINVAL;
830 }
831 break;
832 }
833
Randy Dunlape73a75f2007-07-15 23:40:52 -0700834 default:
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -0700835 printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
836 p);
837 return -EINVAL;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700838 break;
839 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 }
Andi Kleena137e1c2008-07-23 21:27:43 -0700841
842 /* Do size after hstate is set up */
843 if (setsize > NO_SIZE) {
844 struct hstate *h = pconfig->hstate;
845 if (setsize == SIZE_PERCENT) {
846 size <<= huge_page_shift(h);
847 size *= h->max_huge_pages;
848 do_div(size, 100);
849 }
850 pconfig->nr_blocks = (size >> huge_page_shift(h));
851 }
852
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 return 0;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700854
855bad_val:
856 printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
857 args[0].from, p);
Akinobu Mitac12ddba2009-04-21 12:24:05 -0700858 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859}
860
861static int
862hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
863{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 int ret;
865 struct hugetlbfs_config config;
866 struct hugetlbfs_sb_info *sbinfo;
867
Miklos Szeredi10f19a82008-02-08 04:21:45 -0800868 save_mount_options(sb, data);
869
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 config.nr_blocks = -1; /* No limit on size by default */
871 config.nr_inodes = -1; /* No limit on number of inodes by default */
David Howells77c70de2008-11-14 10:38:56 +1100872 config.uid = current_fsuid();
873 config.gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 config.mode = 0755;
Andi Kleena137e1c2008-07-23 21:27:43 -0700875 config.hstate = &default_hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 ret = hugetlbfs_parse_options(data, &config);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 if (ret)
878 return ret;
879
880 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
881 if (!sbinfo)
882 return -ENOMEM;
883 sb->s_fs_info = sbinfo;
Andi Kleena137e1c2008-07-23 21:27:43 -0700884 sbinfo->hstate = config.hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 spin_lock_init(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 sbinfo->max_inodes = config.nr_inodes;
887 sbinfo->free_inodes = config.nr_inodes;
David Gibson90481622012-03-21 16:34:12 -0700888 sbinfo->spool = NULL;
889 if (config.nr_blocks != -1) {
890 sbinfo->spool = hugepage_new_subpool(config.nr_blocks);
891 if (!sbinfo->spool)
892 goto out_free;
893 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 sb->s_maxbytes = MAX_LFS_FILESIZE;
Andi Kleena137e1c2008-07-23 21:27:43 -0700895 sb->s_blocksize = huge_page_size(config.hstate);
896 sb->s_blocksize_bits = huge_page_shift(config.hstate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 sb->s_magic = HUGETLBFS_MAGIC;
898 sb->s_op = &hugetlbfs_ops;
899 sb->s_time_gran = 1;
Al Viro48fde702012-01-08 22:15:13 -0500900 sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
901 if (!sb->s_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 return 0;
904out_free:
David Gibson90481622012-03-21 16:34:12 -0700905 if (sbinfo->spool)
906 kfree(sbinfo->spool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 kfree(sbinfo);
908 return -ENOMEM;
909}
910
Al Viro3c26ff62010-07-25 11:46:36 +0400911static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
912 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
Al Viro3c26ff62010-07-25 11:46:36 +0400914 return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915}
916
917static struct file_system_type hugetlbfs_fs_type = {
918 .name = "hugetlbfs",
Al Viro3c26ff62010-07-25 11:46:36 +0400919 .mount = hugetlbfs_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 .kill_sb = kill_litter_super,
921};
922
923static struct vfsmount *hugetlbfs_vfsmount;
924
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700925static int can_do_hugetlb_shm(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926{
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700927 return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
Steven Truelove40716e22012-03-21 16:34:14 -0700930struct file *hugetlb_file_setup(const char *name, unsigned long addr,
931 size_t size, vm_flags_t acctflag,
Eric B Munson6bfde052009-09-21 17:03:43 -0700932 struct user_struct **user, int creat_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933{
934 int error = -ENOMEM;
935 struct file *file;
936 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400937 struct path path;
938 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 struct qstr quick_string;
Steven Truelove40716e22012-03-21 16:34:14 -0700940 struct hstate *hstate;
941 unsigned long num_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Hugh Dickins353d5c32009-08-24 16:30:28 +0100943 *user = NULL;
Akinobu Mita5bc98592007-05-06 14:50:18 -0700944 if (!hugetlbfs_vfsmount)
945 return ERR_PTR(-ENOENT);
946
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700947 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
Hugh Dickins353d5c32009-08-24 16:30:28 +0100948 *user = current_user();
949 if (user_shm_lock(size, *user)) {
David Rientjes21a3c272012-03-21 16:34:13 -0700950 task_lock(current);
951 printk_once(KERN_WARNING
952 "%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
953 current->comm, current->pid);
954 task_unlock(current);
Hugh Dickins353d5c32009-08-24 16:30:28 +0100955 } else {
956 *user = NULL;
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -0700957 return ERR_PTR(-EPERM);
Hugh Dickins353d5c32009-08-24 16:30:28 +0100958 }
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -0700959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 root = hugetlbfs_vfsmount->mnt_root;
Eric W. Biederman9d665862007-06-16 10:16:16 -0700962 quick_string.name = name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 quick_string.len = strlen(quick_string.name);
964 quick_string.hash = 0;
Al Viro2c48b9c2009-08-09 00:52:35 +0400965 path.dentry = d_alloc(root, &quick_string);
966 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 goto out_shm_unlock;
968
Al Viro2c48b9c2009-08-09 00:52:35 +0400969 path.mnt = mntget(hugetlbfs_vfsmount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 error = -ENOSPC;
Al Viro7d54fa62011-07-24 20:20:48 -0400971 inode = hugetlbfs_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 if (!inode)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700973 goto out_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
Steven Truelove40716e22012-03-21 16:34:14 -0700975 hstate = hstate_inode(inode);
976 size += addr & ~huge_page_mask(hstate);
977 num_pages = ALIGN(size, huge_page_size(hstate)) >>
978 huge_page_shift(hstate);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800979 error = -ENOMEM;
Steven Truelove40716e22012-03-21 16:34:14 -0700980 if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag))
David Gibsonb45b5bd2006-03-22 00:08:55 -0800981 goto out_inode;
982
Al Viro2c48b9c2009-08-09 00:52:35 +0400983 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +0200985 clear_nlink(inode);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700986
987 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +0400988 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Dave Hansence8d2cd2007-10-16 23:31:13 -0700989 &hugetlbfs_file_operations);
990 if (!file)
Al Virob4d232e2008-02-23 05:59:19 -0500991 goto out_dentry; /* inode is already attached */
Dave Hansence8d2cd2007-10-16 23:31:13 -0700992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 return file;
994
David Gibsonb45b5bd2006-03-22 00:08:55 -0800995out_inode:
996 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997out_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +0400998 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999out_shm_unlock:
Hugh Dickins353d5c32009-08-24 16:30:28 +01001000 if (*user) {
1001 user_shm_unlock(size, *user);
1002 *user = NULL;
1003 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 return ERR_PTR(error);
1005}
1006
1007static int __init init_hugetlbfs_fs(void)
1008{
1009 int error;
1010 struct vfsmount *vfsmount;
1011
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001012 error = bdi_init(&hugetlbfs_backing_dev_info);
1013 if (error)
1014 return error;
1015
Hillf Dantond1d5e052012-03-21 16:34:15 -07001016 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1018 sizeof(struct hugetlbfs_inode_info),
Paul Mundt20c2df82007-07-20 10:11:58 +09001019 0, 0, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 if (hugetlbfs_inode_cachep == NULL)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001021 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
1023 error = register_filesystem(&hugetlbfs_fs_type);
1024 if (error)
1025 goto out;
1026
1027 vfsmount = kern_mount(&hugetlbfs_fs_type);
1028
1029 if (!IS_ERR(vfsmount)) {
1030 hugetlbfs_vfsmount = vfsmount;
1031 return 0;
1032 }
1033
1034 error = PTR_ERR(vfsmount);
1035
1036 out:
Hillf Dantond1d5e052012-03-21 16:34:15 -07001037 kmem_cache_destroy(hugetlbfs_inode_cachep);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001038 out2:
1039 bdi_destroy(&hugetlbfs_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 return error;
1041}
1042
1043static void __exit exit_hugetlbfs_fs(void)
1044{
1045 kmem_cache_destroy(hugetlbfs_inode_cachep);
Tim Chen423e0ab2011-07-19 09:32:38 -07001046 kern_unmount(hugetlbfs_vfsmount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 unregister_filesystem(&hugetlbfs_fs_type);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001048 bdi_destroy(&hugetlbfs_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049}
1050
1051module_init(init_hugetlbfs_fs)
1052module_exit(exit_hugetlbfs_fs)
1053
1054MODULE_LICENSE("GPL");