blob: 7d479ce3aceb602c687a7f022a7d3d6c1284de0f [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/quotaops.h>
30#include <linux/slab.h>
31#include <linux/dnotify.h>
32#include <linux/statfs.h>
33#include <linux/security.h>
34
35#include <asm/uaccess.h>
36
37/* some random number */
38#define HUGETLBFS_MAGIC 0x958458f6
39
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -080040static const struct super_operations hugetlbfs_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070041static const struct address_space_operations hugetlbfs_aops;
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080042const struct file_operations hugetlbfs_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080043static const struct inode_operations hugetlbfs_dir_inode_operations;
44static const struct inode_operations hugetlbfs_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46static struct backing_dev_info hugetlbfs_backing_dev_info = {
47 .ra_pages = 0, /* No readahead */
Miklos Szeredie4ad08f2008-04-30 00:54:37 -070048 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Linus Torvalds1da177e2005-04-16 15:20:36 -070049};
50
51int sysctl_hugetlb_shm_group;
52
Randy Dunlape73a75f2007-07-15 23:40:52 -070053enum {
54 Opt_size, Opt_nr_inodes,
55 Opt_mode, Opt_uid, Opt_gid,
Andi Kleena137e1c2008-07-23 21:27:43 -070056 Opt_pagesize,
Randy Dunlape73a75f2007-07-15 23:40:52 -070057 Opt_err,
58};
59
Steven Whitehousea447c092008-10-13 10:46:57 +010060static const match_table_t tokens = {
Randy Dunlape73a75f2007-07-15 23:40:52 -070061 {Opt_size, "size=%s"},
62 {Opt_nr_inodes, "nr_inodes=%s"},
63 {Opt_mode, "mode=%o"},
64 {Opt_uid, "uid=%u"},
65 {Opt_gid, "gid=%u"},
Andi Kleena137e1c2008-07-23 21:27:43 -070066 {Opt_pagesize, "pagesize=%s"},
Randy Dunlape73a75f2007-07-15 23:40:52 -070067 {Opt_err, NULL},
68};
69
Adam Litke2e9b3672005-10-29 18:16:47 -070070static void huge_pagevec_release(struct pagevec *pvec)
71{
72 int i;
73
74 for (i = 0; i < pagevec_count(pvec); ++i)
75 put_page(pvec->pages[i]);
76
77 pagevec_reinit(pvec);
78}
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
81{
Josef Sipekb39424e2006-12-08 02:37:07 -080082 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 loff_t len, vma_len;
84 int ret;
Andi Kleena5516432008-07-23 21:27:41 -070085 struct hstate *h = hstate_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Hugh Dickins68589bc2006-11-14 02:03:32 -080087 /*
David Gibsondec4ad82007-08-30 23:56:40 -070088 * vma address alignment (but not the pgoff alignment) has
89 * already been checked by prepare_hugepage_range. If you add
90 * any error returns here, do so after setting VM_HUGETLB, so
91 * is_vm_hugetlb_page tests below unmap_region go the right
92 * way when do_mmap_pgoff unwinds (may be important on powerpc
93 * and ia64).
Hugh Dickins68589bc2006-11-14 02:03:32 -080094 */
95 vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
96 vma->vm_ops = &hugetlb_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Andi Kleena5516432008-07-23 21:27:41 -070098 if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
David Gibsondec4ad82007-08-30 23:56:40 -070099 return -EINVAL;
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
102
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800103 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 file_accessed(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
106 ret = -ENOMEM;
107 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
Mel Gormana1e78772008-07-23 21:27:23 -0700109 if (hugetlb_reserve_pages(inode,
Andi Kleena5516432008-07-23 21:27:41 -0700110 vma->vm_pgoff >> huge_page_order(h),
111 len >> huge_page_shift(h), vma))
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;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700255
256 /* nr is the maximum number of bytes to copy from this page */
Andi Kleena5516432008-07-23 21:27:41 -0700257 nr = huge_page_size(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700258 if (index >= end_index) {
259 if (index > end_index)
260 goto out;
Andi Kleena5516432008-07-23 21:27:41 -0700261 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700262 if (nr <= offset) {
263 goto out;
264 }
265 }
266 nr = nr - offset;
267
268 /* Find the page */
269 page = find_get_page(mapping, index);
270 if (unlikely(page == NULL)) {
271 /*
272 * We have a HOLE, zero out the user-buffer for the
273 * length of the hole or request.
274 */
275 ret = len < nr ? len : nr;
276 if (clear_user(buf, ret))
277 ret = -EFAULT;
278 } else {
279 /*
280 * We have the page, copy it to user space buffer.
281 */
282 ret = hugetlbfs_read_actor(page, offset, buf, len, nr);
283 }
284 if (ret < 0) {
285 if (retval == 0)
286 retval = ret;
287 if (page)
288 page_cache_release(page);
289 goto out;
290 }
291
292 offset += ret;
293 retval += ret;
294 len -= ret;
Andi Kleena5516432008-07-23 21:27:41 -0700295 index += offset >> huge_page_shift(h);
296 offset &= ~huge_page_mask(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700297
298 if (page)
299 page_cache_release(page);
300
301 /* short read or no more work */
302 if ((ret != nr) || (len == 0))
303 break;
304 }
305out:
Andi Kleena5516432008-07-23 21:27:41 -0700306 *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700307 mutex_unlock(&inode->i_mutex);
308 return retval;
309}
310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311/*
312 * Read a page. Again trivial. If it didn't already exist
313 * in the page cache, it is zero-filled.
314 */
315static int hugetlbfs_readpage(struct file *file, struct page * page)
316{
317 unlock_page(page);
318 return -EINVAL;
319}
320
Nick Piggin800d15a2007-10-16 01:25:03 -0700321static int hugetlbfs_write_begin(struct file *file,
322 struct address_space *mapping,
323 loff_t pos, unsigned len, unsigned flags,
324 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 return -EINVAL;
327}
328
Nick Piggin800d15a2007-10-16 01:25:03 -0700329static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
330 loff_t pos, unsigned len, unsigned copied,
331 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
Nick Piggin800d15a2007-10-16 01:25:03 -0700333 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 return -EINVAL;
335}
336
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337static void truncate_huge_page(struct page *page)
338{
Linus Torvaldsfba25912006-12-20 13:46:42 -0800339 cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 ClearPageUptodate(page);
341 remove_from_page_cache(page);
342 put_page(page);
343}
344
David Gibsonb45b5bd2006-03-22 00:08:55 -0800345static void truncate_hugepages(struct inode *inode, loff_t lstart)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
Andi Kleena5516432008-07-23 21:27:41 -0700347 struct hstate *h = hstate_inode(inode);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800348 struct address_space *mapping = &inode->i_data;
Andi Kleena5516432008-07-23 21:27:41 -0700349 const pgoff_t start = lstart >> huge_page_shift(h);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 struct pagevec pvec;
351 pgoff_t next;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700352 int i, freed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354 pagevec_init(&pvec, 0);
355 next = start;
356 while (1) {
357 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
358 if (next == start)
359 break;
360 next = start;
361 continue;
362 }
363
364 for (i = 0; i < pagevec_count(&pvec); ++i) {
365 struct page *page = pvec.pages[i];
366
367 lock_page(page);
368 if (page->index > next)
369 next = page->index;
370 ++next;
371 truncate_huge_page(page);
372 unlock_page(page);
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700373 freed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 }
375 huge_pagevec_release(&pvec);
376 }
377 BUG_ON(!lstart && mapping->nrpages);
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700378 hugetlb_unreserve_pages(inode, start, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379}
380
381static void hugetlbfs_delete_inode(struct inode *inode)
382{
David Gibsonb45b5bd2006-03-22 00:08:55 -0800383 truncate_hugepages(inode, 0);
Christoph Hellwig149f4212005-10-29 18:16:43 -0700384 clear_inode(inode);
385}
386
Josh Triplettddc0a512006-09-29 01:59:27 -0700387static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
Christoph Hellwig0b1533f2005-10-29 18:16:45 -0700389 struct super_block *sb = inode->i_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
Christoph Hellwig0b1533f2005-10-29 18:16:45 -0700391 if (!hlist_unhashed(&inode->i_hash)) {
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700392 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Christoph Hellwig0b1533f2005-10-29 18:16:45 -0700393 list_move(&inode->i_list, &inode_unused);
394 inodes_stat.nr_unused++;
395 if (!sb || (sb->s_flags & MS_ACTIVE)) {
396 spin_unlock(&inode_lock);
397 return;
398 }
399 inode->i_state |= I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 spin_unlock(&inode_lock);
Christoph Hellwig0b1533f2005-10-29 18:16:45 -0700401 /*
402 * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
403 * in our backing_dev_info.
404 */
405 write_inode_now(inode, 1);
406 spin_lock(&inode_lock);
407 inode->i_state &= ~I_WILL_FREE;
408 inodes_stat.nr_unused--;
409 hlist_del_init(&inode->i_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 list_del_init(&inode->i_list);
412 list_del_init(&inode->i_sb_list);
413 inode->i_state |= I_FREEING;
414 inodes_stat.nr_inodes--;
415 spin_unlock(&inode_lock);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800416 truncate_hugepages(inode, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 clear_inode(inode);
418 destroy_inode(inode);
419}
420
421static void hugetlbfs_drop_inode(struct inode *inode)
422{
423 if (!inode->i_nlink)
Christoph Hellwig6b09b9d2005-10-29 18:16:44 -0700424 generic_delete_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 else
426 hugetlbfs_forget_inode(inode);
427}
428
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429static inline void
Hugh Dickins856fc292006-10-28 10:38:43 -0700430hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
432 struct vm_area_struct *vma;
433 struct prio_tree_iter iter;
434
Hugh Dickins856fc292006-10-28 10:38:43 -0700435 vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 unsigned long v_offset;
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 /*
Hugh Dickins856fc292006-10-28 10:38:43 -0700439 * Can the expression below overflow on 32-bit arches?
440 * No, because the prio_tree returns us only those vmas
441 * which overlap the truncated area starting at pgoff,
442 * and no vma on a 32-bit arch can span beyond the 4GB.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 */
Hugh Dickins856fc292006-10-28 10:38:43 -0700444 if (vma->vm_pgoff < pgoff)
445 v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
446 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 v_offset = 0;
448
Chen, Kenneth W502717f2006-10-11 01:20:46 -0700449 __unmap_hugepage_range(vma,
Mel Gorman04f2cbe2008-07-23 21:27:25 -0700450 vma->vm_start + v_offset, vma->vm_end, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 }
452}
453
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
455{
Hugh Dickins856fc292006-10-28 10:38:43 -0700456 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 struct address_space *mapping = inode->i_mapping;
Andi Kleena5516432008-07-23 21:27:41 -0700458 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Andi Kleena5516432008-07-23 21:27:41 -0700460 BUG_ON(offset & ~huge_page_mask(h));
Hugh Dickins856fc292006-10-28 10:38:43 -0700461 pgoff = offset >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Ken Chen7aa91e12007-10-16 01:26:21 -0700463 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 spin_lock(&mapping->i_mmap_lock);
465 if (!prio_tree_empty(&mapping->i_mmap))
466 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
467 spin_unlock(&mapping->i_mmap_lock);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800468 truncate_hugepages(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 return 0;
470}
471
472static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
473{
474 struct inode *inode = dentry->d_inode;
Andi Kleena5516432008-07-23 21:27:41 -0700475 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 int error;
477 unsigned int ia_valid = attr->ia_valid;
478
479 BUG_ON(!inode);
480
481 error = inode_change_ok(inode, attr);
482 if (error)
483 goto out;
484
485 if (ia_valid & ATTR_SIZE) {
486 error = -EINVAL;
Andi Kleena5516432008-07-23 21:27:41 -0700487 if (!(attr->ia_size & ~huge_page_mask(h)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 error = hugetlb_vmtruncate(inode, attr->ia_size);
489 if (error)
490 goto out;
491 attr->ia_valid &= ~ATTR_SIZE;
492 }
493 error = inode_setattr(inode, attr);
494out:
495 return error;
496}
497
498static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
499 gid_t gid, int mode, dev_t dev)
500{
501 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
503 inode = new_inode(sb);
504 if (inode) {
505 struct hugetlbfs_inode_info *info;
506 inode->i_mode = mode;
507 inode->i_uid = uid;
508 inode->i_gid = gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 inode->i_blocks = 0;
510 inode->i_mapping->a_ops = &hugetlbfs_aops;
511 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
512 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700513 INIT_LIST_HEAD(&inode->i_mapping->private_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 info = HUGETLBFS_I(inode);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700515 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 switch (mode & S_IFMT) {
517 default:
518 init_special_inode(inode, mode, dev);
519 break;
520 case S_IFREG:
521 inode->i_op = &hugetlbfs_inode_operations;
522 inode->i_fop = &hugetlbfs_file_operations;
523 break;
524 case S_IFDIR:
525 inode->i_op = &hugetlbfs_dir_inode_operations;
526 inode->i_fop = &simple_dir_operations;
527
528 /* directory inodes start off with i_nlink == 2 (for "." entry) */
Dave Hansend8c76e62006-09-30 23:29:04 -0700529 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 break;
531 case S_IFLNK:
532 inode->i_op = &page_symlink_inode_operations;
533 break;
534 }
535 }
536 return inode;
537}
538
539/*
540 * File creation. Allocate an inode, and we're done..
541 */
542static int hugetlbfs_mknod(struct inode *dir,
543 struct dentry *dentry, int mode, dev_t dev)
544{
545 struct inode *inode;
546 int error = -ENOSPC;
547 gid_t gid;
548
549 if (dir->i_mode & S_ISGID) {
550 gid = dir->i_gid;
551 if (S_ISDIR(mode))
552 mode |= S_ISGID;
553 } else {
David Howells77c70de2008-11-14 10:38:56 +1100554 gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 }
David Howells77c70de2008-11-14 10:38:56 +1100556 inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 if (inode) {
558 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
559 d_instantiate(dentry, inode);
560 dget(dentry); /* Extra count - pin the dentry in core */
561 error = 0;
562 }
563 return error;
564}
565
566static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
567{
568 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
569 if (!retval)
Dave Hansend8c76e62006-09-30 23:29:04 -0700570 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 return retval;
572}
573
574static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
575{
576 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
577}
578
579static int hugetlbfs_symlink(struct inode *dir,
580 struct dentry *dentry, const char *symname)
581{
582 struct inode *inode;
583 int error = -ENOSPC;
584 gid_t gid;
585
586 if (dir->i_mode & S_ISGID)
587 gid = dir->i_gid;
588 else
David Howells77c70de2008-11-14 10:38:56 +1100589 gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
David Howells77c70de2008-11-14 10:38:56 +1100591 inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 gid, S_IFLNK|S_IRWXUGO, 0);
593 if (inode) {
594 int l = strlen(symname)+1;
595 error = page_symlink(inode, symname, l);
596 if (!error) {
597 d_instantiate(dentry, inode);
598 dget(dentry);
599 } else
600 iput(inode);
601 }
602 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
603
604 return error;
605}
606
607/*
Ken Chen6649a382007-02-08 14:20:27 -0800608 * mark the head page dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 */
610static int hugetlbfs_set_page_dirty(struct page *page)
611{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700612 struct page *head = compound_head(page);
Ken Chen6649a382007-02-08 14:20:27 -0800613
614 SetPageDirty(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 return 0;
616}
617
David Howells726c3342006-06-23 02:02:58 -0700618static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
David Howells726c3342006-06-23 02:02:58 -0700620 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
Andi Kleena5516432008-07-23 21:27:41 -0700621 struct hstate *h = hstate_inode(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
623 buf->f_type = HUGETLBFS_MAGIC;
Andi Kleena5516432008-07-23 21:27:41 -0700624 buf->f_bsize = huge_page_size(h);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 if (sbinfo) {
626 spin_lock(&sbinfo->stat_lock);
David Gibson74a8a652005-11-21 21:32:24 -0800627 /* If no limits set, just report 0 for max/free/used
628 * blocks, like simple_statfs() */
629 if (sbinfo->max_blocks >= 0) {
630 buf->f_blocks = sbinfo->max_blocks;
631 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
632 buf->f_files = sbinfo->max_inodes;
633 buf->f_ffree = sbinfo->free_inodes;
634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 spin_unlock(&sbinfo->stat_lock);
636 }
637 buf->f_namelen = NAME_MAX;
638 return 0;
639}
640
641static void hugetlbfs_put_super(struct super_block *sb)
642{
643 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
644
645 if (sbi) {
646 sb->s_fs_info = NULL;
647 kfree(sbi);
648 }
649}
650
Christoph Hellwig96527982005-10-29 18:16:42 -0700651static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
652{
653 if (sbinfo->free_inodes >= 0) {
654 spin_lock(&sbinfo->stat_lock);
655 if (unlikely(!sbinfo->free_inodes)) {
656 spin_unlock(&sbinfo->stat_lock);
657 return 0;
658 }
659 sbinfo->free_inodes--;
660 spin_unlock(&sbinfo->stat_lock);
661 }
662
663 return 1;
664}
665
666static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
667{
668 if (sbinfo->free_inodes >= 0) {
669 spin_lock(&sbinfo->stat_lock);
670 sbinfo->free_inodes++;
671 spin_unlock(&sbinfo->stat_lock);
672 }
673}
674
675
Christoph Lametere18b8902006-12-06 20:33:20 -0800676static struct kmem_cache *hugetlbfs_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
678static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
679{
Christoph Hellwig96527982005-10-29 18:16:42 -0700680 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 struct hugetlbfs_inode_info *p;
682
Christoph Hellwig96527982005-10-29 18:16:42 -0700683 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 return NULL;
Christoph Lametere94b1762006-12-06 20:33:17 -0800685 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
Christoph Hellwig96527982005-10-29 18:16:42 -0700686 if (unlikely(!p)) {
687 hugetlbfs_inc_free_inodes(sbinfo);
688 return NULL;
689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 return &p->vfs_inode;
691}
692
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693static void hugetlbfs_destroy_inode(struct inode *inode)
694{
Christoph Hellwig96527982005-10-29 18:16:42 -0700695 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
697 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
698}
699
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700700static const struct address_space_operations hugetlbfs_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 .readpage = hugetlbfs_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -0700702 .write_begin = hugetlbfs_write_begin,
703 .write_end = hugetlbfs_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 .set_page_dirty = hugetlbfs_set_page_dirty,
705};
706
Christoph Hellwig96527982005-10-29 18:16:42 -0700707
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700708static void init_once(void *foo)
Christoph Hellwig96527982005-10-29 18:16:42 -0700709{
710 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
711
Christoph Lametera35afb82007-05-16 22:10:57 -0700712 inode_init_once(&ei->vfs_inode);
Christoph Hellwig96527982005-10-29 18:16:42 -0700713}
714
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800715const struct file_operations hugetlbfs_file_operations = {
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700716 .read = hugetlbfs_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 .mmap = hugetlbfs_file_mmap,
718 .fsync = simple_sync_file,
719 .get_unmapped_area = hugetlb_get_unmapped_area,
720};
721
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800722static const struct inode_operations hugetlbfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 .create = hugetlbfs_create,
724 .lookup = simple_lookup,
725 .link = simple_link,
726 .unlink = simple_unlink,
727 .symlink = hugetlbfs_symlink,
728 .mkdir = hugetlbfs_mkdir,
729 .rmdir = simple_rmdir,
730 .mknod = hugetlbfs_mknod,
731 .rename = simple_rename,
732 .setattr = hugetlbfs_setattr,
733};
734
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800735static const struct inode_operations hugetlbfs_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 .setattr = hugetlbfs_setattr,
737};
738
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800739static const struct super_operations hugetlbfs_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 .alloc_inode = hugetlbfs_alloc_inode,
741 .destroy_inode = hugetlbfs_destroy_inode,
742 .statfs = hugetlbfs_statfs,
Christoph Hellwig149f4212005-10-29 18:16:43 -0700743 .delete_inode = hugetlbfs_delete_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 .drop_inode = hugetlbfs_drop_inode,
745 .put_super = hugetlbfs_put_super,
Miklos Szeredi10f19a82008-02-08 04:21:45 -0800746 .show_options = generic_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747};
748
749static int
750hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
751{
Randy Dunlape73a75f2007-07-15 23:40:52 -0700752 char *p, *rest;
753 substring_t args[MAX_OPT_ARGS];
754 int option;
Andi Kleena137e1c2008-07-23 21:27:43 -0700755 unsigned long long size = 0;
756 enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
758 if (!options)
759 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Randy Dunlape73a75f2007-07-15 23:40:52 -0700761 while ((p = strsep(&options, ",")) != NULL) {
762 int token;
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -0700763 if (!*p)
764 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Randy Dunlape73a75f2007-07-15 23:40:52 -0700766 token = match_token(p, tokens, args);
767 switch (token) {
768 case Opt_uid:
769 if (match_int(&args[0], &option))
770 goto bad_val;
771 pconfig->uid = option;
772 break;
773
774 case Opt_gid:
775 if (match_int(&args[0], &option))
776 goto bad_val;
777 pconfig->gid = option;
778 break;
779
780 case Opt_mode:
781 if (match_octal(&args[0], &option))
782 goto bad_val;
Ken Chen75897d62008-02-04 22:28:36 -0800783 pconfig->mode = option & 01777U;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700784 break;
785
786 case Opt_size: {
Randy Dunlape73a75f2007-07-15 23:40:52 -0700787 /* memparse() will accept a K/M/G without a digit */
788 if (!isdigit(*args[0].from))
789 goto bad_val;
790 size = memparse(args[0].from, &rest);
Andi Kleena137e1c2008-07-23 21:27:43 -0700791 setsize = SIZE_STD;
792 if (*rest == '%')
793 setsize = SIZE_PERCENT;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700794 break;
795 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Randy Dunlape73a75f2007-07-15 23:40:52 -0700797 case Opt_nr_inodes:
798 /* memparse() will accept a K/M/G without a digit */
799 if (!isdigit(*args[0].from))
800 goto bad_val;
801 pconfig->nr_inodes = memparse(args[0].from, &rest);
802 break;
803
Andi Kleena137e1c2008-07-23 21:27:43 -0700804 case Opt_pagesize: {
805 unsigned long ps;
806 ps = memparse(args[0].from, &rest);
807 pconfig->hstate = size_to_hstate(ps);
808 if (!pconfig->hstate) {
809 printk(KERN_ERR
810 "hugetlbfs: Unsupported page size %lu MB\n",
811 ps >> 20);
812 return -EINVAL;
813 }
814 break;
815 }
816
Randy Dunlape73a75f2007-07-15 23:40:52 -0700817 default:
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -0700818 printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
819 p);
820 return -EINVAL;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700821 break;
822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 }
Andi Kleena137e1c2008-07-23 21:27:43 -0700824
825 /* Do size after hstate is set up */
826 if (setsize > NO_SIZE) {
827 struct hstate *h = pconfig->hstate;
828 if (setsize == SIZE_PERCENT) {
829 size <<= huge_page_shift(h);
830 size *= h->max_huge_pages;
831 do_div(size, 100);
832 }
833 pconfig->nr_blocks = (size >> huge_page_shift(h));
834 }
835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 return 0;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700837
838bad_val:
839 printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
840 args[0].from, p);
841 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842}
843
844static int
845hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
846{
847 struct inode * inode;
848 struct dentry * root;
849 int ret;
850 struct hugetlbfs_config config;
851 struct hugetlbfs_sb_info *sbinfo;
852
Miklos Szeredi10f19a82008-02-08 04:21:45 -0800853 save_mount_options(sb, data);
854
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 config.nr_blocks = -1; /* No limit on size by default */
856 config.nr_inodes = -1; /* No limit on number of inodes by default */
David Howells77c70de2008-11-14 10:38:56 +1100857 config.uid = current_fsuid();
858 config.gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 config.mode = 0755;
Andi Kleena137e1c2008-07-23 21:27:43 -0700860 config.hstate = &default_hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 ret = hugetlbfs_parse_options(data, &config);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 if (ret)
863 return ret;
864
865 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
866 if (!sbinfo)
867 return -ENOMEM;
868 sb->s_fs_info = sbinfo;
Andi Kleena137e1c2008-07-23 21:27:43 -0700869 sbinfo->hstate = config.hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 spin_lock_init(&sbinfo->stat_lock);
871 sbinfo->max_blocks = config.nr_blocks;
872 sbinfo->free_blocks = config.nr_blocks;
873 sbinfo->max_inodes = config.nr_inodes;
874 sbinfo->free_inodes = config.nr_inodes;
875 sb->s_maxbytes = MAX_LFS_FILESIZE;
Andi Kleena137e1c2008-07-23 21:27:43 -0700876 sb->s_blocksize = huge_page_size(config.hstate);
877 sb->s_blocksize_bits = huge_page_shift(config.hstate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 sb->s_magic = HUGETLBFS_MAGIC;
879 sb->s_op = &hugetlbfs_ops;
880 sb->s_time_gran = 1;
881 inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
882 S_IFDIR | config.mode, 0);
883 if (!inode)
884 goto out_free;
885
886 root = d_alloc_root(inode);
887 if (!root) {
888 iput(inode);
889 goto out_free;
890 }
891 sb->s_root = root;
892 return 0;
893out_free:
894 kfree(sbinfo);
895 return -ENOMEM;
896}
897
Adam Litke9a119c02007-11-14 16:59:41 -0800898int hugetlb_get_quota(struct address_space *mapping, long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899{
900 int ret = 0;
901 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
902
903 if (sbinfo->free_blocks > -1) {
904 spin_lock(&sbinfo->stat_lock);
Adam Litke9a119c02007-11-14 16:59:41 -0800905 if (sbinfo->free_blocks - delta >= 0)
906 sbinfo->free_blocks -= delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 else
908 ret = -ENOMEM;
909 spin_unlock(&sbinfo->stat_lock);
910 }
911
912 return ret;
913}
914
Adam Litke9a119c02007-11-14 16:59:41 -0800915void hugetlb_put_quota(struct address_space *mapping, long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916{
917 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
918
919 if (sbinfo->free_blocks > -1) {
920 spin_lock(&sbinfo->stat_lock);
Adam Litke9a119c02007-11-14 16:59:41 -0800921 sbinfo->free_blocks += delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 spin_unlock(&sbinfo->stat_lock);
923 }
924}
925
David Howells454e2392006-06-23 02:02:57 -0700926static int hugetlbfs_get_sb(struct file_system_type *fs_type,
927 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
David Howells454e2392006-06-23 02:02:57 -0700929 return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930}
931
932static struct file_system_type hugetlbfs_fs_type = {
933 .name = "hugetlbfs",
934 .get_sb = hugetlbfs_get_sb,
935 .kill_sb = kill_litter_super,
936};
937
938static struct vfsmount *hugetlbfs_vfsmount;
939
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940static int can_do_hugetlb_shm(void)
941{
942 return likely(capable(CAP_IPC_LOCK) ||
943 in_group_p(sysctl_hugetlb_shm_group) ||
944 can_do_mlock());
945}
946
Eric W. Biederman9d665862007-06-16 10:16:16 -0700947struct file *hugetlb_file_setup(const char *name, size_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948{
949 int error = -ENOMEM;
950 struct file *file;
951 struct inode *inode;
952 struct dentry *dentry, *root;
953 struct qstr quick_string;
David Howells86a264a2008-11-14 10:39:18 +1100954 struct user_struct *user = current_user();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Akinobu Mita5bc98592007-05-06 14:50:18 -0700956 if (!hugetlbfs_vfsmount)
957 return ERR_PTR(-ENOENT);
958
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 if (!can_do_hugetlb_shm())
960 return ERR_PTR(-EPERM);
961
David Howells86a264a2008-11-14 10:39:18 +1100962 if (!user_shm_lock(size, user))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 return ERR_PTR(-ENOMEM);
964
965 root = hugetlbfs_vfsmount->mnt_root;
Eric W. Biederman9d665862007-06-16 10:16:16 -0700966 quick_string.name = name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 quick_string.len = strlen(quick_string.name);
968 quick_string.hash = 0;
969 dentry = d_alloc(root, &quick_string);
970 if (!dentry)
971 goto out_shm_unlock;
972
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 error = -ENOSPC;
David Howells77c70de2008-11-14 10:38:56 +1100974 inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
975 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 if (!inode)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700977 goto out_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
David Gibsonb45b5bd2006-03-22 00:08:55 -0800979 error = -ENOMEM;
Andi Kleena5516432008-07-23 21:27:41 -0700980 if (hugetlb_reserve_pages(inode, 0,
981 size >> huge_page_shift(hstate_inode(inode)), NULL))
David Gibsonb45b5bd2006-03-22 00:08:55 -0800982 goto out_inode;
983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 d_instantiate(dentry, inode);
985 inode->i_size = size;
986 inode->i_nlink = 0;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700987
988 error = -ENFILE;
989 file = alloc_file(hugetlbfs_vfsmount, dentry,
990 FMODE_WRITE | FMODE_READ,
991 &hugetlbfs_file_operations);
992 if (!file)
Al Virob4d232e2008-02-23 05:59:19 -0500993 goto out_dentry; /* inode is already attached */
Dave Hansence8d2cd2007-10-16 23:31:13 -0700994
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 return file;
996
David Gibsonb45b5bd2006-03-22 00:08:55 -0800997out_inode:
998 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999out_dentry:
1000 dput(dentry);
1001out_shm_unlock:
David Howells86a264a2008-11-14 10:39:18 +11001002 user_shm_unlock(size, user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 return ERR_PTR(error);
1004}
1005
1006static int __init init_hugetlbfs_fs(void)
1007{
1008 int error;
1009 struct vfsmount *vfsmount;
1010
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001011 error = bdi_init(&hugetlbfs_backing_dev_info);
1012 if (error)
1013 return error;
1014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1016 sizeof(struct hugetlbfs_inode_info),
Paul Mundt20c2df82007-07-20 10:11:58 +09001017 0, 0, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 if (hugetlbfs_inode_cachep == NULL)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001019 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
1021 error = register_filesystem(&hugetlbfs_fs_type);
1022 if (error)
1023 goto out;
1024
1025 vfsmount = kern_mount(&hugetlbfs_fs_type);
1026
1027 if (!IS_ERR(vfsmount)) {
1028 hugetlbfs_vfsmount = vfsmount;
1029 return 0;
1030 }
1031
1032 error = PTR_ERR(vfsmount);
1033
1034 out:
1035 if (error)
1036 kmem_cache_destroy(hugetlbfs_inode_cachep);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001037 out2:
1038 bdi_destroy(&hugetlbfs_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 return error;
1040}
1041
1042static void __exit exit_hugetlbfs_fs(void)
1043{
1044 kmem_cache_destroy(hugetlbfs_inode_cachep);
1045 unregister_filesystem(&hugetlbfs_fs_type);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001046 bdi_destroy(&hugetlbfs_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047}
1048
1049module_init(init_hugetlbfs_fs)
1050module_exit(exit_hugetlbfs_fs)
1051
1052MODULE_LICENSE("GPL");