blob: 0be5a78598d02a71f627ac0acb0ec83365b41845 [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
44static struct backing_dev_info hugetlbfs_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +020045 .name = "hugetlbfs",
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 .ra_pages = 0, /* No readahead */
Miklos Szeredie4ad08f2008-04-30 00:54:37 -070047 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Linus Torvalds1da177e2005-04-16 15:20:36 -070048};
49
50int sysctl_hugetlb_shm_group;
51
Randy Dunlape73a75f2007-07-15 23:40:52 -070052enum {
53 Opt_size, Opt_nr_inodes,
54 Opt_mode, Opt_uid, Opt_gid,
Andi Kleena137e1c2008-07-23 21:27:43 -070055 Opt_pagesize,
Randy Dunlape73a75f2007-07-15 23:40:52 -070056 Opt_err,
57};
58
Steven Whitehousea447c092008-10-13 10:46:57 +010059static const match_table_t tokens = {
Randy Dunlape73a75f2007-07-15 23:40:52 -070060 {Opt_size, "size=%s"},
61 {Opt_nr_inodes, "nr_inodes=%s"},
62 {Opt_mode, "mode=%o"},
63 {Opt_uid, "uid=%u"},
64 {Opt_gid, "gid=%u"},
Andi Kleena137e1c2008-07-23 21:27:43 -070065 {Opt_pagesize, "pagesize=%s"},
Randy Dunlape73a75f2007-07-15 23:40:52 -070066 {Opt_err, NULL},
67};
68
Adam Litke2e9b3672005-10-29 18:16:47 -070069static void huge_pagevec_release(struct pagevec *pvec)
70{
71 int i;
72
73 for (i = 0; i < pagevec_count(pvec); ++i)
74 put_page(pvec->pages[i]);
75
76 pagevec_reinit(pvec);
77}
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
80{
Josef Sipekb39424e2006-12-08 02:37:07 -080081 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 loff_t len, vma_len;
83 int ret;
Andi Kleena5516432008-07-23 21:27:41 -070084 struct hstate *h = hstate_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Hugh Dickins68589bc2006-11-14 02:03:32 -080086 /*
David Gibsondec4ad82007-08-30 23:56:40 -070087 * vma address alignment (but not the pgoff alignment) has
88 * already been checked by prepare_hugepage_range. If you add
89 * any error returns here, do so after setting VM_HUGETLB, so
90 * is_vm_hugetlb_page tests below unmap_region go the right
91 * way when do_mmap_pgoff unwinds (may be important on powerpc
92 * and ia64).
Hugh Dickins68589bc2006-11-14 02:03:32 -080093 */
94 vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
95 vma->vm_ops = &hugetlb_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
Becky Bruce2b37c352011-07-25 17:11:49 -070097 if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
David Gibsondec4ad82007-08-30 23:56:40 -070098 return -EINVAL;
99
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
101
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800102 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 file_accessed(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
105 ret = -ENOMEM;
106 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Mel Gormana1e78772008-07-23 21:27:23 -0700108 if (hugetlb_reserve_pages(inode,
Andi Kleena5516432008-07-23 21:27:41 -0700109 vma->vm_pgoff >> huge_page_order(h),
Mel Gorman5a6fe122009-02-10 14:02:27 +0000110 len >> huge_page_shift(h), vma,
111 vma->vm_flags))
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700112 goto out;
David Gibsonb45b5bd2006-03-22 00:08:55 -0800113
Adam Litke4c887262005-10-29 18:16:46 -0700114 ret = 0;
115 hugetlb_prefault_arch_hook(vma->vm_mm);
Zhang, Yanminb6174df2006-07-10 04:44:49 -0700116 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 inode->i_size = len;
118out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800119 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
121 return ret;
122}
123
124/*
Hugh Dickins508034a2005-10-29 18:16:30 -0700125 * Called under down_write(mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 */
127
Adrian Bunkd2ba27e82007-05-06 14:49:00 -0700128#ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129static unsigned long
130hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
131 unsigned long len, unsigned long pgoff, unsigned long flags)
132{
133 struct mm_struct *mm = current->mm;
134 struct vm_area_struct *vma;
135 unsigned long start_addr;
Andi Kleena5516432008-07-23 21:27:41 -0700136 struct hstate *h = hstate_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
Andi Kleena5516432008-07-23 21:27:41 -0700138 if (len & ~huge_page_mask(h))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 return -EINVAL;
140 if (len > TASK_SIZE)
141 return -ENOMEM;
142
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -0700143 if (flags & MAP_FIXED) {
Andi Kleena5516432008-07-23 21:27:41 -0700144 if (prepare_hugepage_range(file, addr, len))
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -0700145 return -EINVAL;
146 return addr;
147 }
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 if (addr) {
Andi Kleena5516432008-07-23 21:27:41 -0700150 addr = ALIGN(addr, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 vma = find_vma(mm, addr);
152 if (TASK_SIZE - len >= addr &&
153 (!vma || addr + len <= vma->vm_start))
154 return addr;
155 }
156
157 start_addr = mm->free_area_cache;
158
Wolfgang Wander1363c3c2005-06-21 17:14:49 -0700159 if (len <= mm->cached_hole_size)
160 start_addr = TASK_UNMAPPED_BASE;
161
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162full_search:
Andi Kleena5516432008-07-23 21:27:41 -0700163 addr = ALIGN(start_addr, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
166 /* At this point: (!vma || addr < vma->vm_end). */
167 if (TASK_SIZE - len < addr) {
168 /*
169 * Start a new search - just in case we missed
170 * some holes.
171 */
172 if (start_addr != TASK_UNMAPPED_BASE) {
173 start_addr = TASK_UNMAPPED_BASE;
174 goto full_search;
175 }
176 return -ENOMEM;
177 }
178
179 if (!vma || addr + len <= vma->vm_start)
180 return addr;
Andi Kleena5516432008-07-23 21:27:41 -0700181 addr = ALIGN(vma->vm_end, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 }
183}
184#endif
185
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700186static int
187hugetlbfs_read_actor(struct page *page, unsigned long offset,
188 char __user *buf, unsigned long count,
189 unsigned long size)
190{
191 char *kaddr;
192 unsigned long left, copied = 0;
193 int i, chunksize;
194
195 if (size > count)
196 size = count;
197
198 /* Find which 4k chunk and offset with in that chunk */
199 i = offset >> PAGE_CACHE_SHIFT;
200 offset = offset & ~PAGE_CACHE_MASK;
201
202 while (size) {
203 chunksize = PAGE_CACHE_SIZE;
204 if (offset)
205 chunksize -= offset;
206 if (chunksize > size)
207 chunksize = size;
208 kaddr = kmap(&page[i]);
209 left = __copy_to_user(buf, kaddr + offset, chunksize);
210 kunmap(&page[i]);
211 if (left) {
212 copied += (chunksize - left);
213 break;
214 }
215 offset = 0;
216 size -= chunksize;
217 buf += chunksize;
218 copied += chunksize;
219 i++;
220 }
221 return copied ? copied : -EFAULT;
222}
223
224/*
225 * Support for read() - Find the page attached to f_mapping and copy out the
226 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
227 * since it has PAGE_CACHE_SIZE assumptions.
228 */
229static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
230 size_t len, loff_t *ppos)
231{
Andi Kleena5516432008-07-23 21:27:41 -0700232 struct hstate *h = hstate_file(filp);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700233 struct address_space *mapping = filp->f_mapping;
234 struct inode *inode = mapping->host;
Andi Kleena5516432008-07-23 21:27:41 -0700235 unsigned long index = *ppos >> huge_page_shift(h);
236 unsigned long offset = *ppos & ~huge_page_mask(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700237 unsigned long end_index;
238 loff_t isize;
239 ssize_t retval = 0;
240
241 mutex_lock(&inode->i_mutex);
242
243 /* validate length */
244 if (len == 0)
245 goto out;
246
247 isize = i_size_read(inode);
248 if (!isize)
249 goto out;
250
Andi Kleena5516432008-07-23 21:27:41 -0700251 end_index = (isize - 1) >> huge_page_shift(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700252 for (;;) {
253 struct page *page;
Andi Kleena5516432008-07-23 21:27:41 -0700254 unsigned long nr, ret;
Roel Kluin91bf189c2009-01-06 14:40:14 -0800255 int ra;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700256
257 /* nr is the maximum number of bytes to copy from this page */
Andi Kleena5516432008-07-23 21:27:41 -0700258 nr = huge_page_size(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700259 if (index >= end_index) {
260 if (index > end_index)
261 goto out;
Andi Kleena5516432008-07-23 21:27:41 -0700262 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700263 if (nr <= offset) {
264 goto out;
265 }
266 }
267 nr = nr - offset;
268
269 /* Find the page */
270 page = find_get_page(mapping, index);
271 if (unlikely(page == NULL)) {
272 /*
273 * We have a HOLE, zero out the user-buffer for the
274 * length of the hole or request.
275 */
276 ret = len < nr ? len : nr;
277 if (clear_user(buf, ret))
Roel Kluin91bf189c2009-01-06 14:40:14 -0800278 ra = -EFAULT;
279 else
280 ra = 0;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700281 } else {
282 /*
283 * We have the page, copy it to user space buffer.
284 */
Roel Kluin91bf189c2009-01-06 14:40:14 -0800285 ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
286 ret = ra;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700287 }
Roel Kluin91bf189c2009-01-06 14:40:14 -0800288 if (ra < 0) {
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700289 if (retval == 0)
Roel Kluin91bf189c2009-01-06 14:40:14 -0800290 retval = ra;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700291 if (page)
292 page_cache_release(page);
293 goto out;
294 }
295
296 offset += ret;
297 retval += ret;
298 len -= ret;
Andi Kleena5516432008-07-23 21:27:41 -0700299 index += offset >> huge_page_shift(h);
300 offset &= ~huge_page_mask(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700301
302 if (page)
303 page_cache_release(page);
304
305 /* short read or no more work */
306 if ((ret != nr) || (len == 0))
307 break;
308 }
309out:
Andi Kleena5516432008-07-23 21:27:41 -0700310 *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700311 mutex_unlock(&inode->i_mutex);
312 return retval;
313}
314
Nick Piggin800d15a2007-10-16 01:25:03 -0700315static int hugetlbfs_write_begin(struct file *file,
316 struct address_space *mapping,
317 loff_t pos, unsigned len, unsigned flags,
318 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
320 return -EINVAL;
321}
322
Nick Piggin800d15a2007-10-16 01:25:03 -0700323static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
324 loff_t pos, unsigned len, unsigned copied,
325 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326{
Nick Piggin800d15a2007-10-16 01:25:03 -0700327 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 return -EINVAL;
329}
330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331static void truncate_huge_page(struct page *page)
332{
Linus Torvaldsfba25912006-12-20 13:46:42 -0800333 cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 ClearPageUptodate(page);
Minchan Kimbd65cb82011-03-22 16:30:54 -0700335 delete_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336}
337
David Gibsonb45b5bd2006-03-22 00:08:55 -0800338static void truncate_hugepages(struct inode *inode, loff_t lstart)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339{
Andi Kleena5516432008-07-23 21:27:41 -0700340 struct hstate *h = hstate_inode(inode);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800341 struct address_space *mapping = &inode->i_data;
Andi Kleena5516432008-07-23 21:27:41 -0700342 const pgoff_t start = lstart >> huge_page_shift(h);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 struct pagevec pvec;
344 pgoff_t next;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700345 int i, freed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
347 pagevec_init(&pvec, 0);
348 next = start;
349 while (1) {
350 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
351 if (next == start)
352 break;
353 next = start;
354 continue;
355 }
356
357 for (i = 0; i < pagevec_count(&pvec); ++i) {
358 struct page *page = pvec.pages[i];
359
360 lock_page(page);
361 if (page->index > next)
362 next = page->index;
363 ++next;
364 truncate_huge_page(page);
365 unlock_page(page);
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700366 freed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 }
368 huge_pagevec_release(&pvec);
369 }
370 BUG_ON(!lstart && mapping->nrpages);
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700371 hugetlb_unreserve_pages(inode, start, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372}
373
Al Viro2bbbda32010-06-04 19:52:12 -0400374static void hugetlbfs_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
David Gibsonb45b5bd2006-03-22 00:08:55 -0800376 truncate_hugepages(inode, 0);
Al Virob0683aa2010-06-04 20:55:25 -0400377 end_writeback(inode);
Christoph Hellwig149f4212005-10-29 18:16:43 -0700378}
379
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380static inline void
Hugh Dickins856fc292006-10-28 10:38:43 -0700381hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382{
383 struct vm_area_struct *vma;
384 struct prio_tree_iter iter;
385
Hugh Dickins856fc292006-10-28 10:38:43 -0700386 vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 unsigned long v_offset;
388
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 /*
Hugh Dickins856fc292006-10-28 10:38:43 -0700390 * Can the expression below overflow on 32-bit arches?
391 * No, because the prio_tree returns us only those vmas
392 * which overlap the truncated area starting at pgoff,
393 * and no vma on a 32-bit arch can span beyond the 4GB.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 */
Hugh Dickins856fc292006-10-28 10:38:43 -0700395 if (vma->vm_pgoff < pgoff)
396 v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
397 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 v_offset = 0;
399
Chen, Kenneth W502717f2006-10-11 01:20:46 -0700400 __unmap_hugepage_range(vma,
Mel Gorman04f2cbe2008-07-23 21:27:25 -0700401 vma->vm_start + v_offset, vma->vm_end, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 }
403}
404
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
406{
Hugh Dickins856fc292006-10-28 10:38:43 -0700407 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 struct address_space *mapping = inode->i_mapping;
Andi Kleena5516432008-07-23 21:27:41 -0700409 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Andi Kleena5516432008-07-23 21:27:41 -0700411 BUG_ON(offset & ~huge_page_mask(h));
Hugh Dickins856fc292006-10-28 10:38:43 -0700412 pgoff = offset >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Ken Chen7aa91e12007-10-16 01:26:21 -0700414 i_size_write(inode, offset);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700415 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 if (!prio_tree_empty(&mapping->i_mmap))
417 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700418 mutex_unlock(&mapping->i_mmap_mutex);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800419 truncate_hugepages(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 return 0;
421}
422
423static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
424{
425 struct inode *inode = dentry->d_inode;
Andi Kleena5516432008-07-23 21:27:41 -0700426 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 int error;
428 unsigned int ia_valid = attr->ia_valid;
429
430 BUG_ON(!inode);
431
432 error = inode_change_ok(inode, attr);
433 if (error)
Christoph Hellwig10257742010-06-04 11:30:02 +0200434 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
436 if (ia_valid & ATTR_SIZE) {
437 error = -EINVAL;
Christoph Hellwig10257742010-06-04 11:30:02 +0200438 if (attr->ia_size & ~huge_page_mask(h))
439 return -EINVAL;
440 error = hugetlb_vmtruncate(inode, attr->ia_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 if (error)
Christoph Hellwig10257742010-06-04 11:30:02 +0200442 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 }
Christoph Hellwig10257742010-06-04 11:30:02 +0200444
445 setattr_copy(inode, attr);
446 mark_inode_dirty(inode);
447 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
450static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
451 gid_t gid, int mode, dev_t dev)
452{
453 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455 inode = new_inode(sb);
456 if (inode) {
457 struct hugetlbfs_inode_info *info;
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400458 inode->i_ino = get_next_ino();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 inode->i_mode = mode;
460 inode->i_uid = uid;
461 inode->i_gid = gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 inode->i_mapping->a_ops = &hugetlbfs_aops;
463 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
464 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700465 INIT_LIST_HEAD(&inode->i_mapping->private_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 info = HUGETLBFS_I(inode);
Eric B Munson6bfde052009-09-21 17:03:43 -0700467 /*
468 * The policy is initialized here even if we are creating a
469 * private inode because initialization simply creates an
470 * an empty rb tree and calls spin_lock_init(), later when we
471 * call mpol_free_shared_policy() it will just return because
472 * the rb tree will still be empty.
473 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700474 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 switch (mode & S_IFMT) {
476 default:
477 init_special_inode(inode, mode, dev);
478 break;
479 case S_IFREG:
480 inode->i_op = &hugetlbfs_inode_operations;
481 inode->i_fop = &hugetlbfs_file_operations;
482 break;
483 case S_IFDIR:
484 inode->i_op = &hugetlbfs_dir_inode_operations;
485 inode->i_fop = &simple_dir_operations;
486
487 /* directory inodes start off with i_nlink == 2 (for "." entry) */
Dave Hansend8c76e62006-09-30 23:29:04 -0700488 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 break;
490 case S_IFLNK:
491 inode->i_op = &page_symlink_inode_operations;
492 break;
493 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400494 lockdep_annotate_inode_mutex_key(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
496 return inode;
497}
498
499/*
500 * File creation. Allocate an inode, and we're done..
501 */
502static int hugetlbfs_mknod(struct inode *dir,
503 struct dentry *dentry, int mode, dev_t dev)
504{
505 struct inode *inode;
506 int error = -ENOSPC;
507 gid_t gid;
508
509 if (dir->i_mode & S_ISGID) {
510 gid = dir->i_gid;
511 if (S_ISDIR(mode))
512 mode |= S_ISGID;
513 } else {
David Howells77c70de2008-11-14 10:38:56 +1100514 gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 }
David Howells77c70de2008-11-14 10:38:56 +1100516 inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 if (inode) {
518 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
519 d_instantiate(dentry, inode);
520 dget(dentry); /* Extra count - pin the dentry in core */
521 error = 0;
522 }
523 return error;
524}
525
526static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
527{
528 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
529 if (!retval)
Dave Hansend8c76e62006-09-30 23:29:04 -0700530 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 return retval;
532}
533
534static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
535{
536 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
537}
538
539static int hugetlbfs_symlink(struct inode *dir,
540 struct dentry *dentry, const char *symname)
541{
542 struct inode *inode;
543 int error = -ENOSPC;
544 gid_t gid;
545
546 if (dir->i_mode & S_ISGID)
547 gid = dir->i_gid;
548 else
David Howells77c70de2008-11-14 10:38:56 +1100549 gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
David Howells77c70de2008-11-14 10:38:56 +1100551 inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 gid, S_IFLNK|S_IRWXUGO, 0);
553 if (inode) {
554 int l = strlen(symname)+1;
555 error = page_symlink(inode, symname, l);
556 if (!error) {
557 d_instantiate(dentry, inode);
558 dget(dentry);
559 } else
560 iput(inode);
561 }
562 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
563
564 return error;
565}
566
567/*
Ken Chen6649a382007-02-08 14:20:27 -0800568 * mark the head page dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 */
570static int hugetlbfs_set_page_dirty(struct page *page)
571{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700572 struct page *head = compound_head(page);
Ken Chen6649a382007-02-08 14:20:27 -0800573
574 SetPageDirty(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 return 0;
576}
577
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900578static int hugetlbfs_migrate_page(struct address_space *mapping,
579 struct page *newpage, struct page *page)
580{
581 int rc;
582
583 rc = migrate_huge_page_move_mapping(mapping, newpage, page);
584 if (rc)
585 return rc;
586 migrate_page_copy(newpage, page);
587
588 return 0;
589}
590
David Howells726c3342006-06-23 02:02:58 -0700591static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592{
David Howells726c3342006-06-23 02:02:58 -0700593 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
Andi Kleena5516432008-07-23 21:27:41 -0700594 struct hstate *h = hstate_inode(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
596 buf->f_type = HUGETLBFS_MAGIC;
Andi Kleena5516432008-07-23 21:27:41 -0700597 buf->f_bsize = huge_page_size(h);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 if (sbinfo) {
599 spin_lock(&sbinfo->stat_lock);
David Gibson74a8a652005-11-21 21:32:24 -0800600 /* If no limits set, just report 0 for max/free/used
601 * blocks, like simple_statfs() */
602 if (sbinfo->max_blocks >= 0) {
603 buf->f_blocks = sbinfo->max_blocks;
604 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
605 buf->f_files = sbinfo->max_inodes;
606 buf->f_ffree = sbinfo->free_inodes;
607 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 spin_unlock(&sbinfo->stat_lock);
609 }
610 buf->f_namelen = NAME_MAX;
611 return 0;
612}
613
614static void hugetlbfs_put_super(struct super_block *sb)
615{
616 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
617
618 if (sbi) {
619 sb->s_fs_info = NULL;
620 kfree(sbi);
621 }
622}
623
Christoph Hellwig96527982005-10-29 18:16:42 -0700624static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
625{
626 if (sbinfo->free_inodes >= 0) {
627 spin_lock(&sbinfo->stat_lock);
628 if (unlikely(!sbinfo->free_inodes)) {
629 spin_unlock(&sbinfo->stat_lock);
630 return 0;
631 }
632 sbinfo->free_inodes--;
633 spin_unlock(&sbinfo->stat_lock);
634 }
635
636 return 1;
637}
638
639static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
640{
641 if (sbinfo->free_inodes >= 0) {
642 spin_lock(&sbinfo->stat_lock);
643 sbinfo->free_inodes++;
644 spin_unlock(&sbinfo->stat_lock);
645 }
646}
647
648
Christoph Lametere18b8902006-12-06 20:33:20 -0800649static struct kmem_cache *hugetlbfs_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
651static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
652{
Christoph Hellwig96527982005-10-29 18:16:42 -0700653 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 struct hugetlbfs_inode_info *p;
655
Christoph Hellwig96527982005-10-29 18:16:42 -0700656 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 return NULL;
Christoph Lametere94b1762006-12-06 20:33:17 -0800658 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
Christoph Hellwig96527982005-10-29 18:16:42 -0700659 if (unlikely(!p)) {
660 hugetlbfs_inc_free_inodes(sbinfo);
661 return NULL;
662 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 return &p->vfs_inode;
664}
665
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100666static void hugetlbfs_i_callback(struct rcu_head *head)
667{
668 struct inode *inode = container_of(head, struct inode, i_rcu);
669 INIT_LIST_HEAD(&inode->i_dentry);
670 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
671}
672
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673static void hugetlbfs_destroy_inode(struct inode *inode)
674{
Christoph Hellwig96527982005-10-29 18:16:42 -0700675 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100677 call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678}
679
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700680static const struct address_space_operations hugetlbfs_aops = {
Nick Piggin800d15a2007-10-16 01:25:03 -0700681 .write_begin = hugetlbfs_write_begin,
682 .write_end = hugetlbfs_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 .set_page_dirty = hugetlbfs_set_page_dirty,
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900684 .migratepage = hugetlbfs_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685};
686
Christoph Hellwig96527982005-10-29 18:16:42 -0700687
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700688static void init_once(void *foo)
Christoph Hellwig96527982005-10-29 18:16:42 -0700689{
690 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
691
Christoph Lametera35afb82007-05-16 22:10:57 -0700692 inode_init_once(&ei->vfs_inode);
Christoph Hellwig96527982005-10-29 18:16:42 -0700693}
694
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800695const struct file_operations hugetlbfs_file_operations = {
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700696 .read = hugetlbfs_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 .mmap = hugetlbfs_file_mmap,
Christoph Hellwig1b061d92010-05-26 17:53:41 +0200698 .fsync = noop_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 .get_unmapped_area = hugetlb_get_unmapped_area,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200700 .llseek = default_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701};
702
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800703static const struct inode_operations hugetlbfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 .create = hugetlbfs_create,
705 .lookup = simple_lookup,
706 .link = simple_link,
707 .unlink = simple_unlink,
708 .symlink = hugetlbfs_symlink,
709 .mkdir = hugetlbfs_mkdir,
710 .rmdir = simple_rmdir,
711 .mknod = hugetlbfs_mknod,
712 .rename = simple_rename,
713 .setattr = hugetlbfs_setattr,
714};
715
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800716static const struct inode_operations hugetlbfs_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 .setattr = hugetlbfs_setattr,
718};
719
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800720static const struct super_operations hugetlbfs_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 .alloc_inode = hugetlbfs_alloc_inode,
722 .destroy_inode = hugetlbfs_destroy_inode,
Al Viro2bbbda32010-06-04 19:52:12 -0400723 .evict_inode = hugetlbfs_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 .statfs = hugetlbfs_statfs,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 .put_super = hugetlbfs_put_super,
Miklos Szeredi10f19a82008-02-08 04:21:45 -0800726 .show_options = generic_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727};
728
729static int
730hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
731{
Randy Dunlape73a75f2007-07-15 23:40:52 -0700732 char *p, *rest;
733 substring_t args[MAX_OPT_ARGS];
734 int option;
Andi Kleena137e1c2008-07-23 21:27:43 -0700735 unsigned long long size = 0;
736 enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737
738 if (!options)
739 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Randy Dunlape73a75f2007-07-15 23:40:52 -0700741 while ((p = strsep(&options, ",")) != NULL) {
742 int token;
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -0700743 if (!*p)
744 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
Randy Dunlape73a75f2007-07-15 23:40:52 -0700746 token = match_token(p, tokens, args);
747 switch (token) {
748 case Opt_uid:
749 if (match_int(&args[0], &option))
750 goto bad_val;
751 pconfig->uid = option;
752 break;
753
754 case Opt_gid:
755 if (match_int(&args[0], &option))
756 goto bad_val;
757 pconfig->gid = option;
758 break;
759
760 case Opt_mode:
761 if (match_octal(&args[0], &option))
762 goto bad_val;
Ken Chen75897d62008-02-04 22:28:36 -0800763 pconfig->mode = option & 01777U;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700764 break;
765
766 case Opt_size: {
Randy Dunlape73a75f2007-07-15 23:40:52 -0700767 /* memparse() will accept a K/M/G without a digit */
768 if (!isdigit(*args[0].from))
769 goto bad_val;
770 size = memparse(args[0].from, &rest);
Andi Kleena137e1c2008-07-23 21:27:43 -0700771 setsize = SIZE_STD;
772 if (*rest == '%')
773 setsize = SIZE_PERCENT;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700774 break;
775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
Randy Dunlape73a75f2007-07-15 23:40:52 -0700777 case Opt_nr_inodes:
778 /* memparse() will accept a K/M/G without a digit */
779 if (!isdigit(*args[0].from))
780 goto bad_val;
781 pconfig->nr_inodes = memparse(args[0].from, &rest);
782 break;
783
Andi Kleena137e1c2008-07-23 21:27:43 -0700784 case Opt_pagesize: {
785 unsigned long ps;
786 ps = memparse(args[0].from, &rest);
787 pconfig->hstate = size_to_hstate(ps);
788 if (!pconfig->hstate) {
789 printk(KERN_ERR
790 "hugetlbfs: Unsupported page size %lu MB\n",
791 ps >> 20);
792 return -EINVAL;
793 }
794 break;
795 }
796
Randy Dunlape73a75f2007-07-15 23:40:52 -0700797 default:
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -0700798 printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
799 p);
800 return -EINVAL;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700801 break;
802 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 }
Andi Kleena137e1c2008-07-23 21:27:43 -0700804
805 /* Do size after hstate is set up */
806 if (setsize > NO_SIZE) {
807 struct hstate *h = pconfig->hstate;
808 if (setsize == SIZE_PERCENT) {
809 size <<= huge_page_shift(h);
810 size *= h->max_huge_pages;
811 do_div(size, 100);
812 }
813 pconfig->nr_blocks = (size >> huge_page_shift(h));
814 }
815
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 return 0;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700817
818bad_val:
819 printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
820 args[0].from, p);
Akinobu Mitac12ddba2009-04-21 12:24:05 -0700821 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822}
823
824static int
825hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
826{
827 struct inode * inode;
828 struct dentry * root;
829 int ret;
830 struct hugetlbfs_config config;
831 struct hugetlbfs_sb_info *sbinfo;
832
Miklos Szeredi10f19a82008-02-08 04:21:45 -0800833 save_mount_options(sb, data);
834
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 config.nr_blocks = -1; /* No limit on size by default */
836 config.nr_inodes = -1; /* No limit on number of inodes by default */
David Howells77c70de2008-11-14 10:38:56 +1100837 config.uid = current_fsuid();
838 config.gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 config.mode = 0755;
Andi Kleena137e1c2008-07-23 21:27:43 -0700840 config.hstate = &default_hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 ret = hugetlbfs_parse_options(data, &config);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 if (ret)
843 return ret;
844
845 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
846 if (!sbinfo)
847 return -ENOMEM;
848 sb->s_fs_info = sbinfo;
Andi Kleena137e1c2008-07-23 21:27:43 -0700849 sbinfo->hstate = config.hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 spin_lock_init(&sbinfo->stat_lock);
851 sbinfo->max_blocks = config.nr_blocks;
852 sbinfo->free_blocks = config.nr_blocks;
853 sbinfo->max_inodes = config.nr_inodes;
854 sbinfo->free_inodes = config.nr_inodes;
855 sb->s_maxbytes = MAX_LFS_FILESIZE;
Andi Kleena137e1c2008-07-23 21:27:43 -0700856 sb->s_blocksize = huge_page_size(config.hstate);
857 sb->s_blocksize_bits = huge_page_shift(config.hstate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 sb->s_magic = HUGETLBFS_MAGIC;
859 sb->s_op = &hugetlbfs_ops;
860 sb->s_time_gran = 1;
861 inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
862 S_IFDIR | config.mode, 0);
863 if (!inode)
864 goto out_free;
865
866 root = d_alloc_root(inode);
867 if (!root) {
868 iput(inode);
869 goto out_free;
870 }
871 sb->s_root = root;
872 return 0;
873out_free:
874 kfree(sbinfo);
875 return -ENOMEM;
876}
877
Adam Litke9a119c02007-11-14 16:59:41 -0800878int hugetlb_get_quota(struct address_space *mapping, long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879{
880 int ret = 0;
881 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
882
883 if (sbinfo->free_blocks > -1) {
884 spin_lock(&sbinfo->stat_lock);
Adam Litke9a119c02007-11-14 16:59:41 -0800885 if (sbinfo->free_blocks - delta >= 0)
886 sbinfo->free_blocks -= delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 else
888 ret = -ENOMEM;
889 spin_unlock(&sbinfo->stat_lock);
890 }
891
892 return ret;
893}
894
Adam Litke9a119c02007-11-14 16:59:41 -0800895void hugetlb_put_quota(struct address_space *mapping, long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896{
897 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
898
899 if (sbinfo->free_blocks > -1) {
900 spin_lock(&sbinfo->stat_lock);
Adam Litke9a119c02007-11-14 16:59:41 -0800901 sbinfo->free_blocks += delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 spin_unlock(&sbinfo->stat_lock);
903 }
904}
905
Al Viro3c26ff62010-07-25 11:46:36 +0400906static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
907 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908{
Al Viro3c26ff62010-07-25 11:46:36 +0400909 return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910}
911
912static struct file_system_type hugetlbfs_fs_type = {
913 .name = "hugetlbfs",
Al Viro3c26ff62010-07-25 11:46:36 +0400914 .mount = hugetlbfs_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 .kill_sb = kill_litter_super,
916};
917
918static struct vfsmount *hugetlbfs_vfsmount;
919
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700920static int can_do_hugetlb_shm(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921{
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700922 return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923}
924
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900925struct file *hugetlb_file_setup(const char *name, size_t size,
926 vm_flags_t acctflag,
Eric B Munson6bfde052009-09-21 17:03:43 -0700927 struct user_struct **user, int creat_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
929 int error = -ENOMEM;
930 struct file *file;
931 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400932 struct path path;
933 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 struct qstr quick_string;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
Hugh Dickins353d5c32009-08-24 16:30:28 +0100936 *user = NULL;
Akinobu Mita5bc98592007-05-06 14:50:18 -0700937 if (!hugetlbfs_vfsmount)
938 return ERR_PTR(-ENOENT);
939
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700940 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
Hugh Dickins353d5c32009-08-24 16:30:28 +0100941 *user = current_user();
942 if (user_shm_lock(size, *user)) {
Dave Jones52ca0e82010-11-11 14:05:20 -0800943 printk_once(KERN_WARNING "Using mlock ulimits for SHM_HUGETLB is deprecated\n");
Hugh Dickins353d5c32009-08-24 16:30:28 +0100944 } else {
945 *user = NULL;
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -0700946 return ERR_PTR(-EPERM);
Hugh Dickins353d5c32009-08-24 16:30:28 +0100947 }
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -0700948 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 root = hugetlbfs_vfsmount->mnt_root;
Eric W. Biederman9d665862007-06-16 10:16:16 -0700951 quick_string.name = name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 quick_string.len = strlen(quick_string.name);
953 quick_string.hash = 0;
Al Viro2c48b9c2009-08-09 00:52:35 +0400954 path.dentry = d_alloc(root, &quick_string);
955 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 goto out_shm_unlock;
957
Al Viro2c48b9c2009-08-09 00:52:35 +0400958 path.mnt = mntget(hugetlbfs_vfsmount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 error = -ENOSPC;
David Howells77c70de2008-11-14 10:38:56 +1100960 inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
961 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 if (!inode)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700963 goto out_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
David Gibsonb45b5bd2006-03-22 00:08:55 -0800965 error = -ENOMEM;
Andi Kleena5516432008-07-23 21:27:41 -0700966 if (hugetlb_reserve_pages(inode, 0,
Mel Gorman5a6fe122009-02-10 14:02:27 +0000967 size >> huge_page_shift(hstate_inode(inode)), NULL,
968 acctflag))
David Gibsonb45b5bd2006-03-22 00:08:55 -0800969 goto out_inode;
970
Al Viro2c48b9c2009-08-09 00:52:35 +0400971 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +0200973 clear_nlink(inode);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700974
975 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +0400976 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Dave Hansence8d2cd2007-10-16 23:31:13 -0700977 &hugetlbfs_file_operations);
978 if (!file)
Al Virob4d232e2008-02-23 05:59:19 -0500979 goto out_dentry; /* inode is already attached */
Dave Hansence8d2cd2007-10-16 23:31:13 -0700980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 return file;
982
David Gibsonb45b5bd2006-03-22 00:08:55 -0800983out_inode:
984 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985out_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +0400986 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987out_shm_unlock:
Hugh Dickins353d5c32009-08-24 16:30:28 +0100988 if (*user) {
989 user_shm_unlock(size, *user);
990 *user = NULL;
991 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 return ERR_PTR(error);
993}
994
995static int __init init_hugetlbfs_fs(void)
996{
997 int error;
998 struct vfsmount *vfsmount;
999
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001000 error = bdi_init(&hugetlbfs_backing_dev_info);
1001 if (error)
1002 return error;
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1005 sizeof(struct hugetlbfs_inode_info),
Paul Mundt20c2df82007-07-20 10:11:58 +09001006 0, 0, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 if (hugetlbfs_inode_cachep == NULL)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001008 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010 error = register_filesystem(&hugetlbfs_fs_type);
1011 if (error)
1012 goto out;
1013
1014 vfsmount = kern_mount(&hugetlbfs_fs_type);
1015
1016 if (!IS_ERR(vfsmount)) {
1017 hugetlbfs_vfsmount = vfsmount;
1018 return 0;
1019 }
1020
1021 error = PTR_ERR(vfsmount);
1022
1023 out:
1024 if (error)
1025 kmem_cache_destroy(hugetlbfs_inode_cachep);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001026 out2:
1027 bdi_destroy(&hugetlbfs_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 return error;
1029}
1030
1031static void __exit exit_hugetlbfs_fs(void)
1032{
1033 kmem_cache_destroy(hugetlbfs_inode_cachep);
Tim Chen423e0ab2011-07-19 09:32:38 -07001034 kern_unmount(hugetlbfs_vfsmount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 unregister_filesystem(&hugetlbfs_fs_type);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001036 bdi_destroy(&hugetlbfs_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037}
1038
1039module_init(init_hugetlbfs_fs)
1040module_exit(exit_hugetlbfs_fs)
1041
1042MODULE_LICENSE("GPL");