blob: 78bde32ea9518d3fffe72d24864b6959bb369781 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * hugetlbpage-backed filesystem. Based on ramfs.
3 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01004 * Nadia Yvette Chambers, 2002
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
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 {
Eric W. Biedermana0eb3a02012-02-07 16:19:25 -080045 kuid_t uid;
46 kgid_t gid;
David Gibsona1d776e2012-03-21 16:34:12 -070047 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 */
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -0700113 vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND | VM_DONTDUMP;
Hugh Dickins68589bc2006-11-14 02:03:32 -0800114 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;
Andi Kleena5516432008-07-23 21:27:41 -0700154 struct hstate *h = hstate_file(file);
Michel Lespinasse08659352012-12-11 16:02:00 -0800155 struct vm_unmapped_area_info info;
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
Michel Lespinasse08659352012-12-11 16:02:00 -0800176 info.flags = 0;
177 info.length = len;
178 info.low_limit = TASK_UNMAPPED_BASE;
179 info.high_limit = TASK_SIZE;
180 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
181 info.align_offset = 0;
182 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
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
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700241 /* validate length */
242 if (len == 0)
243 goto out;
244
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700245 for (;;) {
246 struct page *page;
Andi Kleena5516432008-07-23 21:27:41 -0700247 unsigned long nr, ret;
Roel Kluin91bf189c2009-01-06 14:40:14 -0800248 int ra;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700249
250 /* nr is the maximum number of bytes to copy from this page */
Andi Kleena5516432008-07-23 21:27:41 -0700251 nr = huge_page_size(h);
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700252 isize = i_size_read(inode);
253 if (!isize)
254 goto out;
255 end_index = (isize - 1) >> huge_page_shift(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700256 if (index >= end_index) {
257 if (index > end_index)
258 goto out;
Andi Kleena5516432008-07-23 21:27:41 -0700259 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700260 if (nr <= offset)
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700261 goto out;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700262 }
263 nr = nr - offset;
264
265 /* Find the page */
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700266 page = find_lock_page(mapping, index);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700267 if (unlikely(page == NULL)) {
268 /*
269 * We have a HOLE, zero out the user-buffer for the
270 * length of the hole or request.
271 */
272 ret = len < nr ? len : nr;
273 if (clear_user(buf, ret))
Roel Kluin91bf189c2009-01-06 14:40:14 -0800274 ra = -EFAULT;
275 else
276 ra = 0;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700277 } else {
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700278 unlock_page(page);
279
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700280 /*
281 * We have the page, copy it to user space buffer.
282 */
Roel Kluin91bf189c2009-01-06 14:40:14 -0800283 ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
284 ret = ra;
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700285 page_cache_release(page);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700286 }
Roel Kluin91bf189c2009-01-06 14:40:14 -0800287 if (ra < 0) {
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700288 if (retval == 0)
Roel Kluin91bf189c2009-01-06 14:40:14 -0800289 retval = ra;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700290 goto out;
291 }
292
293 offset += ret;
294 retval += ret;
295 len -= ret;
Andi Kleena5516432008-07-23 21:27:41 -0700296 index += offset >> huge_page_shift(h);
297 offset &= ~huge_page_mask(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700298
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700299 /* short read or no more work */
300 if ((ret != nr) || (len == 0))
301 break;
302 }
303out:
Andi Kleena5516432008-07-23 21:27:41 -0700304 *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700305 return retval;
306}
307
Nick Piggin800d15a2007-10-16 01:25:03 -0700308static int hugetlbfs_write_begin(struct file *file,
309 struct address_space *mapping,
310 loff_t pos, unsigned len, unsigned flags,
311 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
313 return -EINVAL;
314}
315
Nick Piggin800d15a2007-10-16 01:25:03 -0700316static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
317 loff_t pos, unsigned len, unsigned copied,
318 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Nick Piggin800d15a2007-10-16 01:25:03 -0700320 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 return -EINVAL;
322}
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324static void truncate_huge_page(struct page *page)
325{
Linus Torvaldsfba25912006-12-20 13:46:42 -0800326 cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 ClearPageUptodate(page);
Minchan Kimbd65cb82011-03-22 16:30:54 -0700328 delete_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329}
330
David Gibsonb45b5bd2006-03-22 00:08:55 -0800331static void truncate_hugepages(struct inode *inode, loff_t lstart)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
Andi Kleena5516432008-07-23 21:27:41 -0700333 struct hstate *h = hstate_inode(inode);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800334 struct address_space *mapping = &inode->i_data;
Andi Kleena5516432008-07-23 21:27:41 -0700335 const pgoff_t start = lstart >> huge_page_shift(h);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 struct pagevec pvec;
337 pgoff_t next;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700338 int i, freed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
340 pagevec_init(&pvec, 0);
341 next = start;
342 while (1) {
343 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
344 if (next == start)
345 break;
346 next = start;
347 continue;
348 }
349
350 for (i = 0; i < pagevec_count(&pvec); ++i) {
351 struct page *page = pvec.pages[i];
352
353 lock_page(page);
354 if (page->index > next)
355 next = page->index;
356 ++next;
357 truncate_huge_page(page);
358 unlock_page(page);
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700359 freed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 }
361 huge_pagevec_release(&pvec);
362 }
363 BUG_ON(!lstart && mapping->nrpages);
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700364 hugetlb_unreserve_pages(inode, start, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365}
366
Al Viro2bbbda32010-06-04 19:52:12 -0400367static void hugetlbfs_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368{
David Gibsonb45b5bd2006-03-22 00:08:55 -0800369 truncate_hugepages(inode, 0);
Jan Karadbd57682012-05-03 14:48:02 +0200370 clear_inode(inode);
Christoph Hellwig149f4212005-10-29 18:16:43 -0700371}
372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373static inline void
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700374hugetlb_vmtruncate_list(struct rb_root *root, pgoff_t pgoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
376 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700378 vma_interval_tree_foreach(vma, root, pgoff, ULONG_MAX) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 unsigned long v_offset;
380
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 /*
Hugh Dickins856fc292006-10-28 10:38:43 -0700382 * Can the expression below overflow on 32-bit arches?
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700383 * No, because the interval tree returns us only those vmas
Hugh Dickins856fc292006-10-28 10:38:43 -0700384 * which overlap the truncated area starting at pgoff,
385 * and no vma on a 32-bit arch can span beyond the 4GB.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 */
Hugh Dickins856fc292006-10-28 10:38:43 -0700387 if (vma->vm_pgoff < pgoff)
388 v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
389 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 v_offset = 0;
391
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -0700392 unmap_hugepage_range(vma, vma->vm_start + v_offset,
393 vma->vm_end, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 }
395}
396
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
398{
Hugh Dickins856fc292006-10-28 10:38:43 -0700399 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 struct address_space *mapping = inode->i_mapping;
Andi Kleena5516432008-07-23 21:27:41 -0700401 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Andi Kleena5516432008-07-23 21:27:41 -0700403 BUG_ON(offset & ~huge_page_mask(h));
Hugh Dickins856fc292006-10-28 10:38:43 -0700404 pgoff = offset >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Ken Chen7aa91e12007-10-16 01:26:21 -0700406 i_size_write(inode, offset);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700407 mutex_lock(&mapping->i_mmap_mutex);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700408 if (!RB_EMPTY_ROOT(&mapping->i_mmap))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700410 mutex_unlock(&mapping->i_mmap_mutex);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800411 truncate_hugepages(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 return 0;
413}
414
415static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
416{
417 struct inode *inode = dentry->d_inode;
Andi Kleena5516432008-07-23 21:27:41 -0700418 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 int error;
420 unsigned int ia_valid = attr->ia_valid;
421
422 BUG_ON(!inode);
423
424 error = inode_change_ok(inode, attr);
425 if (error)
Christoph Hellwig10257742010-06-04 11:30:02 +0200426 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
428 if (ia_valid & ATTR_SIZE) {
429 error = -EINVAL;
Christoph Hellwig10257742010-06-04 11:30:02 +0200430 if (attr->ia_size & ~huge_page_mask(h))
431 return -EINVAL;
432 error = hugetlb_vmtruncate(inode, attr->ia_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 if (error)
Christoph Hellwig10257742010-06-04 11:30:02 +0200434 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 }
Christoph Hellwig10257742010-06-04 11:30:02 +0200436
437 setattr_copy(inode, attr);
438 mark_inode_dirty(inode);
439 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440}
441
Al Viro7d54fa62011-07-24 20:20:48 -0400442static struct inode *hugetlbfs_get_root(struct super_block *sb,
443 struct hugetlbfs_config *config)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444{
445 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
447 inode = new_inode(sb);
448 if (inode) {
449 struct hugetlbfs_inode_info *info;
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400450 inode->i_ino = get_next_ino();
Al Viro7d54fa62011-07-24 20:20:48 -0400451 inode->i_mode = S_IFDIR | config->mode;
452 inode->i_uid = config->uid;
453 inode->i_gid = config->gid;
454 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
455 info = HUGETLBFS_I(inode);
456 mpol_shared_policy_init(&info->policy, NULL);
457 inode->i_op = &hugetlbfs_dir_inode_operations;
458 inode->i_fop = &simple_dir_operations;
459 /* directory inodes start off with i_nlink == 2 (for "." entry) */
460 inc_nlink(inode);
Aneesh Kumar K.V65ed7602012-04-25 16:01:50 -0700461 lockdep_annotate_inode_mutex_key(inode);
Al Viro7d54fa62011-07-24 20:20:48 -0400462 }
463 return inode;
464}
465
466static struct inode *hugetlbfs_get_inode(struct super_block *sb,
467 struct inode *dir,
Al Viro18df2252011-07-24 23:17:40 -0400468 umode_t mode, dev_t dev)
Al Viro7d54fa62011-07-24 20:20:48 -0400469{
470 struct inode *inode;
471
472 inode = new_inode(sb);
473 if (inode) {
474 struct hugetlbfs_inode_info *info;
475 inode->i_ino = get_next_ino();
476 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 inode->i_mapping->a_ops = &hugetlbfs_aops;
478 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
479 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700480 INIT_LIST_HEAD(&inode->i_mapping->private_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 info = HUGETLBFS_I(inode);
Eric B Munson6bfde052009-09-21 17:03:43 -0700482 /*
483 * The policy is initialized here even if we are creating a
484 * private inode because initialization simply creates an
485 * an empty rb tree and calls spin_lock_init(), later when we
486 * call mpol_free_shared_policy() it will just return because
487 * the rb tree will still be empty.
488 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700489 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 switch (mode & S_IFMT) {
491 default:
492 init_special_inode(inode, mode, dev);
493 break;
494 case S_IFREG:
495 inode->i_op = &hugetlbfs_inode_operations;
496 inode->i_fop = &hugetlbfs_file_operations;
497 break;
498 case S_IFDIR:
499 inode->i_op = &hugetlbfs_dir_inode_operations;
500 inode->i_fop = &simple_dir_operations;
501
502 /* directory inodes start off with i_nlink == 2 (for "." entry) */
Dave Hansend8c76e62006-09-30 23:29:04 -0700503 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 break;
505 case S_IFLNK:
506 inode->i_op = &page_symlink_inode_operations;
507 break;
508 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400509 lockdep_annotate_inode_mutex_key(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 }
511 return inode;
512}
513
514/*
515 * File creation. Allocate an inode, and we're done..
516 */
517static int hugetlbfs_mknod(struct inode *dir,
Al Viro1a67aaf2011-07-26 01:52:52 -0400518 struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519{
520 struct inode *inode;
521 int error = -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
Al Viro7d54fa62011-07-24 20:20:48 -0400523 inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (inode) {
525 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
526 d_instantiate(dentry, inode);
527 dget(dentry); /* Extra count - pin the dentry in core */
528 error = 0;
529 }
530 return error;
531}
532
Al Viro18bb1db2011-07-26 01:41:39 -0400533static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534{
535 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
536 if (!retval)
Dave Hansend8c76e62006-09-30 23:29:04 -0700537 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 return retval;
539}
540
Al Viroebfc3b42012-06-10 18:05:36 -0400541static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542{
543 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
544}
545
546static int hugetlbfs_symlink(struct inode *dir,
547 struct dentry *dentry, const char *symname)
548{
549 struct inode *inode;
550 int error = -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Al Viro7d54fa62011-07-24 20:20:48 -0400552 inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 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,
Mel Gormanb969c4a2012-01-12 17:19:34 -0800579 struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800580 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900581{
582 int rc;
583
584 rc = migrate_huge_page_move_mapping(mapping, newpage, page);
Rafael Aquini78bd5202012-12-11 16:02:31 -0800585 if (rc != MIGRATEPAGE_SUCCESS)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900586 return rc;
587 migrate_page_copy(newpage, page);
588
Rafael Aquini78bd5202012-12-11 16:02:31 -0800589 return MIGRATEPAGE_SUCCESS;
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900590}
591
David Howells726c3342006-06-23 02:02:58 -0700592static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593{
David Howells726c3342006-06-23 02:02:58 -0700594 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
Andi Kleena5516432008-07-23 21:27:41 -0700595 struct hstate *h = hstate_inode(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597 buf->f_type = HUGETLBFS_MAGIC;
Andi Kleena5516432008-07-23 21:27:41 -0700598 buf->f_bsize = huge_page_size(h);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 if (sbinfo) {
600 spin_lock(&sbinfo->stat_lock);
David Gibson74a8a652005-11-21 21:32:24 -0800601 /* If no limits set, just report 0 for max/free/used
602 * blocks, like simple_statfs() */
David Gibson90481622012-03-21 16:34:12 -0700603 if (sbinfo->spool) {
604 long free_pages;
605
606 spin_lock(&sbinfo->spool->lock);
607 buf->f_blocks = sbinfo->spool->max_hpages;
608 free_pages = sbinfo->spool->max_hpages
609 - sbinfo->spool->used_hpages;
610 buf->f_bavail = buf->f_bfree = free_pages;
611 spin_unlock(&sbinfo->spool->lock);
David Gibson74a8a652005-11-21 21:32:24 -0800612 buf->f_files = sbinfo->max_inodes;
613 buf->f_ffree = sbinfo->free_inodes;
614 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 spin_unlock(&sbinfo->stat_lock);
616 }
617 buf->f_namelen = NAME_MAX;
618 return 0;
619}
620
621static void hugetlbfs_put_super(struct super_block *sb)
622{
623 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
624
625 if (sbi) {
626 sb->s_fs_info = NULL;
David Gibson90481622012-03-21 16:34:12 -0700627
628 if (sbi->spool)
629 hugepage_put_subpool(sbi->spool);
630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 kfree(sbi);
632 }
633}
634
Christoph Hellwig96527982005-10-29 18:16:42 -0700635static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
636{
637 if (sbinfo->free_inodes >= 0) {
638 spin_lock(&sbinfo->stat_lock);
639 if (unlikely(!sbinfo->free_inodes)) {
640 spin_unlock(&sbinfo->stat_lock);
641 return 0;
642 }
643 sbinfo->free_inodes--;
644 spin_unlock(&sbinfo->stat_lock);
645 }
646
647 return 1;
648}
649
650static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
651{
652 if (sbinfo->free_inodes >= 0) {
653 spin_lock(&sbinfo->stat_lock);
654 sbinfo->free_inodes++;
655 spin_unlock(&sbinfo->stat_lock);
656 }
657}
658
659
Christoph Lametere18b8902006-12-06 20:33:20 -0800660static struct kmem_cache *hugetlbfs_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
662static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
663{
Christoph Hellwig96527982005-10-29 18:16:42 -0700664 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 struct hugetlbfs_inode_info *p;
666
Christoph Hellwig96527982005-10-29 18:16:42 -0700667 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 return NULL;
Christoph Lametere94b1762006-12-06 20:33:17 -0800669 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
Christoph Hellwig96527982005-10-29 18:16:42 -0700670 if (unlikely(!p)) {
671 hugetlbfs_inc_free_inodes(sbinfo);
672 return NULL;
673 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 return &p->vfs_inode;
675}
676
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100677static void hugetlbfs_i_callback(struct rcu_head *head)
678{
679 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100680 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
681}
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683static void hugetlbfs_destroy_inode(struct inode *inode)
684{
Christoph Hellwig96527982005-10-29 18:16:42 -0700685 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100687 call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688}
689
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700690static const struct address_space_operations hugetlbfs_aops = {
Nick Piggin800d15a2007-10-16 01:25:03 -0700691 .write_begin = hugetlbfs_write_begin,
692 .write_end = hugetlbfs_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 .set_page_dirty = hugetlbfs_set_page_dirty,
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900694 .migratepage = hugetlbfs_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695};
696
Christoph Hellwig96527982005-10-29 18:16:42 -0700697
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700698static void init_once(void *foo)
Christoph Hellwig96527982005-10-29 18:16:42 -0700699{
700 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
701
Christoph Lametera35afb82007-05-16 22:10:57 -0700702 inode_init_once(&ei->vfs_inode);
Christoph Hellwig96527982005-10-29 18:16:42 -0700703}
704
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800705const struct file_operations hugetlbfs_file_operations = {
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700706 .read = hugetlbfs_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 .mmap = hugetlbfs_file_mmap,
Christoph Hellwig1b061d92010-05-26 17:53:41 +0200708 .fsync = noop_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 .get_unmapped_area = hugetlb_get_unmapped_area,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200710 .llseek = default_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711};
712
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800713static const struct inode_operations hugetlbfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 .create = hugetlbfs_create,
715 .lookup = simple_lookup,
716 .link = simple_link,
717 .unlink = simple_unlink,
718 .symlink = hugetlbfs_symlink,
719 .mkdir = hugetlbfs_mkdir,
720 .rmdir = simple_rmdir,
721 .mknod = hugetlbfs_mknod,
722 .rename = simple_rename,
723 .setattr = hugetlbfs_setattr,
724};
725
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800726static const struct inode_operations hugetlbfs_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 .setattr = hugetlbfs_setattr,
728};
729
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800730static const struct super_operations hugetlbfs_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 .alloc_inode = hugetlbfs_alloc_inode,
732 .destroy_inode = hugetlbfs_destroy_inode,
Al Viro2bbbda32010-06-04 19:52:12 -0400733 .evict_inode = hugetlbfs_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 .statfs = hugetlbfs_statfs,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 .put_super = hugetlbfs_put_super,
Miklos Szeredi10f19a82008-02-08 04:21:45 -0800736 .show_options = generic_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737};
738
739static int
740hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
741{
Randy Dunlape73a75f2007-07-15 23:40:52 -0700742 char *p, *rest;
743 substring_t args[MAX_OPT_ARGS];
744 int option;
Andi Kleena137e1c2008-07-23 21:27:43 -0700745 unsigned long long size = 0;
746 enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
748 if (!options)
749 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Randy Dunlape73a75f2007-07-15 23:40:52 -0700751 while ((p = strsep(&options, ",")) != NULL) {
752 int token;
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -0700753 if (!*p)
754 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Randy Dunlape73a75f2007-07-15 23:40:52 -0700756 token = match_token(p, tokens, args);
757 switch (token) {
758 case Opt_uid:
759 if (match_int(&args[0], &option))
760 goto bad_val;
Eric W. Biedermana0eb3a02012-02-07 16:19:25 -0800761 pconfig->uid = make_kuid(current_user_ns(), option);
762 if (!uid_valid(pconfig->uid))
763 goto bad_val;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700764 break;
765
766 case Opt_gid:
767 if (match_int(&args[0], &option))
768 goto bad_val;
Eric W. Biedermana0eb3a02012-02-07 16:19:25 -0800769 pconfig->gid = make_kgid(current_user_ns(), option);
770 if (!gid_valid(pconfig->gid))
771 goto bad_val;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700772 break;
773
774 case Opt_mode:
775 if (match_octal(&args[0], &option))
776 goto bad_val;
Ken Chen75897d62008-02-04 22:28:36 -0800777 pconfig->mode = option & 01777U;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700778 break;
779
780 case Opt_size: {
Randy Dunlape73a75f2007-07-15 23:40:52 -0700781 /* memparse() will accept a K/M/G without a digit */
782 if (!isdigit(*args[0].from))
783 goto bad_val;
784 size = memparse(args[0].from, &rest);
Andi Kleena137e1c2008-07-23 21:27:43 -0700785 setsize = SIZE_STD;
786 if (*rest == '%')
787 setsize = SIZE_PERCENT;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700788 break;
789 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Randy Dunlape73a75f2007-07-15 23:40:52 -0700791 case Opt_nr_inodes:
792 /* memparse() will accept a K/M/G without a digit */
793 if (!isdigit(*args[0].from))
794 goto bad_val;
795 pconfig->nr_inodes = memparse(args[0].from, &rest);
796 break;
797
Andi Kleena137e1c2008-07-23 21:27:43 -0700798 case Opt_pagesize: {
799 unsigned long ps;
800 ps = memparse(args[0].from, &rest);
801 pconfig->hstate = size_to_hstate(ps);
802 if (!pconfig->hstate) {
803 printk(KERN_ERR
804 "hugetlbfs: Unsupported page size %lu MB\n",
805 ps >> 20);
806 return -EINVAL;
807 }
808 break;
809 }
810
Randy Dunlape73a75f2007-07-15 23:40:52 -0700811 default:
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -0700812 printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
813 p);
814 return -EINVAL;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700815 break;
816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 }
Andi Kleena137e1c2008-07-23 21:27:43 -0700818
819 /* Do size after hstate is set up */
820 if (setsize > NO_SIZE) {
821 struct hstate *h = pconfig->hstate;
822 if (setsize == SIZE_PERCENT) {
823 size <<= huge_page_shift(h);
824 size *= h->max_huge_pages;
825 do_div(size, 100);
826 }
827 pconfig->nr_blocks = (size >> huge_page_shift(h));
828 }
829
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 return 0;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700831
832bad_val:
833 printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
834 args[0].from, p);
Akinobu Mitac12ddba2009-04-21 12:24:05 -0700835 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836}
837
838static int
839hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
840{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 int ret;
842 struct hugetlbfs_config config;
843 struct hugetlbfs_sb_info *sbinfo;
844
Miklos Szeredi10f19a82008-02-08 04:21:45 -0800845 save_mount_options(sb, data);
846
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 config.nr_blocks = -1; /* No limit on size by default */
848 config.nr_inodes = -1; /* No limit on number of inodes by default */
David Howells77c70de2008-11-14 10:38:56 +1100849 config.uid = current_fsuid();
850 config.gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 config.mode = 0755;
Andi Kleena137e1c2008-07-23 21:27:43 -0700852 config.hstate = &default_hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 ret = hugetlbfs_parse_options(data, &config);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 if (ret)
855 return ret;
856
857 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
858 if (!sbinfo)
859 return -ENOMEM;
860 sb->s_fs_info = sbinfo;
Andi Kleena137e1c2008-07-23 21:27:43 -0700861 sbinfo->hstate = config.hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 spin_lock_init(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 sbinfo->max_inodes = config.nr_inodes;
864 sbinfo->free_inodes = config.nr_inodes;
David Gibson90481622012-03-21 16:34:12 -0700865 sbinfo->spool = NULL;
866 if (config.nr_blocks != -1) {
867 sbinfo->spool = hugepage_new_subpool(config.nr_blocks);
868 if (!sbinfo->spool)
869 goto out_free;
870 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 sb->s_maxbytes = MAX_LFS_FILESIZE;
Andi Kleena137e1c2008-07-23 21:27:43 -0700872 sb->s_blocksize = huge_page_size(config.hstate);
873 sb->s_blocksize_bits = huge_page_shift(config.hstate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 sb->s_magic = HUGETLBFS_MAGIC;
875 sb->s_op = &hugetlbfs_ops;
876 sb->s_time_gran = 1;
Al Viro48fde702012-01-08 22:15:13 -0500877 sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
878 if (!sb->s_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 return 0;
881out_free:
David Gibson90481622012-03-21 16:34:12 -0700882 if (sbinfo->spool)
883 kfree(sbinfo->spool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 kfree(sbinfo);
885 return -ENOMEM;
886}
887
Al Viro3c26ff62010-07-25 11:46:36 +0400888static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
889 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
Al Viro3c26ff62010-07-25 11:46:36 +0400891 return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}
893
894static struct file_system_type hugetlbfs_fs_type = {
895 .name = "hugetlbfs",
Al Viro3c26ff62010-07-25 11:46:36 +0400896 .mount = hugetlbfs_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 .kill_sb = kill_litter_super,
898};
899
Andi Kleen42d73952012-12-11 16:01:34 -0800900static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700902static int can_do_hugetlb_shm(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
Eric W. Biedermana0eb3a02012-02-07 16:19:25 -0800904 kgid_t shm_group;
905 shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
906 return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}
908
Andi Kleen42d73952012-12-11 16:01:34 -0800909static int get_hstate_idx(int page_size_log)
910{
911 struct hstate *h;
912
913 if (!page_size_log)
914 return default_hstate_idx;
915 h = size_to_hstate(1 << page_size_log);
916 if (!h)
917 return -1;
918 return h - hstates;
919}
920
Steven Truelove40716e22012-03-21 16:34:14 -0700921struct file *hugetlb_file_setup(const char *name, unsigned long addr,
922 size_t size, vm_flags_t acctflag,
Andi Kleen42d73952012-12-11 16:01:34 -0800923 struct user_struct **user,
924 int creat_flags, int page_size_log)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925{
926 int error = -ENOMEM;
927 struct file *file;
928 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400929 struct path path;
930 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 struct qstr quick_string;
Steven Truelove40716e22012-03-21 16:34:14 -0700932 struct hstate *hstate;
933 unsigned long num_pages;
Andi Kleen42d73952012-12-11 16:01:34 -0800934 int hstate_idx;
935
936 hstate_idx = get_hstate_idx(page_size_log);
937 if (hstate_idx < 0)
938 return ERR_PTR(-ENODEV);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
Hugh Dickins353d5c32009-08-24 16:30:28 +0100940 *user = NULL;
Andi Kleen42d73952012-12-11 16:01:34 -0800941 if (!hugetlbfs_vfsmount[hstate_idx])
Akinobu Mita5bc98592007-05-06 14:50:18 -0700942 return ERR_PTR(-ENOENT);
943
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700944 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
Hugh Dickins353d5c32009-08-24 16:30:28 +0100945 *user = current_user();
946 if (user_shm_lock(size, *user)) {
David Rientjes21a3c272012-03-21 16:34:13 -0700947 task_lock(current);
948 printk_once(KERN_WARNING
949 "%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
950 current->comm, current->pid);
951 task_unlock(current);
Hugh Dickins353d5c32009-08-24 16:30:28 +0100952 } else {
953 *user = NULL;
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -0700954 return ERR_PTR(-EPERM);
Hugh Dickins353d5c32009-08-24 16:30:28 +0100955 }
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -0700956 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Andi Kleen42d73952012-12-11 16:01:34 -0800958 root = hugetlbfs_vfsmount[hstate_idx]->mnt_root;
Eric W. Biederman9d665862007-06-16 10:16:16 -0700959 quick_string.name = name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 quick_string.len = strlen(quick_string.name);
961 quick_string.hash = 0;
Al Viro2c48b9c2009-08-09 00:52:35 +0400962 path.dentry = d_alloc(root, &quick_string);
963 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 goto out_shm_unlock;
965
Andi Kleen42d73952012-12-11 16:01:34 -0800966 path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 error = -ENOSPC;
Al Viro7d54fa62011-07-24 20:20:48 -0400968 inode = hugetlbfs_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 if (!inode)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700970 goto out_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Steven Truelove40716e22012-03-21 16:34:14 -0700972 hstate = hstate_inode(inode);
973 size += addr & ~huge_page_mask(hstate);
974 num_pages = ALIGN(size, huge_page_size(hstate)) >>
975 huge_page_shift(hstate);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800976 error = -ENOMEM;
Steven Truelove40716e22012-03-21 16:34:14 -0700977 if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag))
David Gibsonb45b5bd2006-03-22 00:08:55 -0800978 goto out_inode;
979
Al Viro2c48b9c2009-08-09 00:52:35 +0400980 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +0200982 clear_nlink(inode);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700983
984 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +0400985 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Dave Hansence8d2cd2007-10-16 23:31:13 -0700986 &hugetlbfs_file_operations);
987 if (!file)
Al Virob4d232e2008-02-23 05:59:19 -0500988 goto out_dentry; /* inode is already attached */
Dave Hansence8d2cd2007-10-16 23:31:13 -0700989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 return file;
991
David Gibsonb45b5bd2006-03-22 00:08:55 -0800992out_inode:
993 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994out_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +0400995 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996out_shm_unlock:
Hugh Dickins353d5c32009-08-24 16:30:28 +0100997 if (*user) {
998 user_shm_unlock(size, *user);
999 *user = NULL;
1000 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 return ERR_PTR(error);
1002}
1003
1004static int __init init_hugetlbfs_fs(void)
1005{
Andi Kleen42d73952012-12-11 16:01:34 -08001006 struct hstate *h;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 int error;
Andi Kleen42d73952012-12-11 16:01:34 -08001008 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001010 error = bdi_init(&hugetlbfs_backing_dev_info);
1011 if (error)
1012 return error;
1013
Hillf Dantond1d5e052012-03-21 16:34:15 -07001014 error = -ENOMEM;
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
Andi Kleen42d73952012-12-11 16:01:34 -08001025 i = 0;
1026 for_each_hstate(h) {
1027 char buf[50];
1028 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
Andi Kleen42d73952012-12-11 16:01:34 -08001030 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1031 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1032 buf);
1033
1034 if (IS_ERR(hugetlbfs_vfsmount[i])) {
1035 pr_err("hugetlb: Cannot mount internal hugetlbfs for "
1036 "page size %uK", ps_kb);
1037 error = PTR_ERR(hugetlbfs_vfsmount[i]);
1038 hugetlbfs_vfsmount[i] = NULL;
1039 }
1040 i++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 }
Andi Kleen42d73952012-12-11 16:01:34 -08001042 /* Non default hstates are optional */
1043 if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1044 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
1046 out:
Hillf Dantond1d5e052012-03-21 16:34:15 -07001047 kmem_cache_destroy(hugetlbfs_inode_cachep);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001048 out2:
1049 bdi_destroy(&hugetlbfs_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 return error;
1051}
1052
1053static void __exit exit_hugetlbfs_fs(void)
1054{
Andi Kleen42d73952012-12-11 16:01:34 -08001055 struct hstate *h;
1056 int i;
1057
1058
Kirill A. Shutemov8c0a8532012-09-26 11:33:07 +10001059 /*
1060 * Make sure all delayed rcu free inodes are flushed before we
1061 * destroy cache.
1062 */
1063 rcu_barrier();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 kmem_cache_destroy(hugetlbfs_inode_cachep);
Andi Kleen42d73952012-12-11 16:01:34 -08001065 i = 0;
1066 for_each_hstate(h)
1067 kern_unmount(hugetlbfs_vfsmount[i++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 unregister_filesystem(&hugetlbfs_fs_type);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001069 bdi_destroy(&hugetlbfs_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070}
1071
1072module_init(init_hugetlbfs_fs)
1073module_exit(exit_hugetlbfs_fs)
1074
1075MODULE_LICENSE("GPL");