blob: a4e9a7ec3691a4066deb8948da3aacd12606b896 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35#include <asm/uaccess.h>
36
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -080037static const struct super_operations hugetlbfs_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070038static const struct address_space_operations hugetlbfs_aops;
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080039const struct file_operations hugetlbfs_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080040static const struct inode_operations hugetlbfs_dir_inode_operations;
41static const struct inode_operations hugetlbfs_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43static struct backing_dev_info hugetlbfs_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +020044 .name = "hugetlbfs",
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 .ra_pages = 0, /* No readahead */
Miklos Szeredie4ad08f2008-04-30 00:54:37 -070046 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Linus Torvalds1da177e2005-04-16 15:20:36 -070047};
48
49int sysctl_hugetlb_shm_group;
50
Randy Dunlape73a75f2007-07-15 23:40:52 -070051enum {
52 Opt_size, Opt_nr_inodes,
53 Opt_mode, Opt_uid, Opt_gid,
Andi Kleena137e1c2008-07-23 21:27:43 -070054 Opt_pagesize,
Randy Dunlape73a75f2007-07-15 23:40:52 -070055 Opt_err,
56};
57
Steven Whitehousea447c092008-10-13 10:46:57 +010058static const match_table_t tokens = {
Randy Dunlape73a75f2007-07-15 23:40:52 -070059 {Opt_size, "size=%s"},
60 {Opt_nr_inodes, "nr_inodes=%s"},
61 {Opt_mode, "mode=%o"},
62 {Opt_uid, "uid=%u"},
63 {Opt_gid, "gid=%u"},
Andi Kleena137e1c2008-07-23 21:27:43 -070064 {Opt_pagesize, "pagesize=%s"},
Randy Dunlape73a75f2007-07-15 23:40:52 -070065 {Opt_err, NULL},
66};
67
Adam Litke2e9b3672005-10-29 18:16:47 -070068static void huge_pagevec_release(struct pagevec *pvec)
69{
70 int i;
71
72 for (i = 0; i < pagevec_count(pvec); ++i)
73 put_page(pvec->pages[i]);
74
75 pagevec_reinit(pvec);
76}
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
79{
Josef Sipekb39424e2006-12-08 02:37:07 -080080 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 loff_t len, vma_len;
82 int ret;
Andi Kleena5516432008-07-23 21:27:41 -070083 struct hstate *h = hstate_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Hugh Dickins68589bc2006-11-14 02:03:32 -080085 /*
David Gibsondec4ad82007-08-30 23:56:40 -070086 * vma address alignment (but not the pgoff alignment) has
87 * already been checked by prepare_hugepage_range. If you add
88 * any error returns here, do so after setting VM_HUGETLB, so
89 * is_vm_hugetlb_page tests below unmap_region go the right
90 * way when do_mmap_pgoff unwinds (may be important on powerpc
91 * and ia64).
Hugh Dickins68589bc2006-11-14 02:03:32 -080092 */
93 vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
94 vma->vm_ops = &hugetlb_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
Andi Kleena5516432008-07-23 21:27:41 -070096 if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
David Gibsondec4ad82007-08-30 23:56:40 -070097 return -EINVAL;
98
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
100
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800101 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 file_accessed(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104 ret = -ENOMEM;
105 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Mel Gormana1e78772008-07-23 21:27:23 -0700107 if (hugetlb_reserve_pages(inode,
Andi Kleena5516432008-07-23 21:27:41 -0700108 vma->vm_pgoff >> huge_page_order(h),
Mel Gorman5a6fe122009-02-10 14:02:27 +0000109 len >> huge_page_shift(h), vma,
110 vma->vm_flags))
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700111 goto out;
David Gibsonb45b5bd2006-03-22 00:08:55 -0800112
Adam Litke4c887262005-10-29 18:16:46 -0700113 ret = 0;
114 hugetlb_prefault_arch_hook(vma->vm_mm);
Zhang, Yanminb6174df2006-07-10 04:44:49 -0700115 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 inode->i_size = len;
117out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800118 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
120 return ret;
121}
122
123/*
Hugh Dickins508034a2005-10-29 18:16:30 -0700124 * Called under down_write(mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 */
126
Adrian Bunkd2ba27e82007-05-06 14:49:00 -0700127#ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128static unsigned long
129hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
130 unsigned long len, unsigned long pgoff, unsigned long flags)
131{
132 struct mm_struct *mm = current->mm;
133 struct vm_area_struct *vma;
134 unsigned long start_addr;
Andi Kleena5516432008-07-23 21:27:41 -0700135 struct hstate *h = hstate_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
Andi Kleena5516432008-07-23 21:27:41 -0700137 if (len & ~huge_page_mask(h))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 return -EINVAL;
139 if (len > TASK_SIZE)
140 return -ENOMEM;
141
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -0700142 if (flags & MAP_FIXED) {
Andi Kleena5516432008-07-23 21:27:41 -0700143 if (prepare_hugepage_range(file, addr, len))
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -0700144 return -EINVAL;
145 return addr;
146 }
147
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 if (addr) {
Andi Kleena5516432008-07-23 21:27:41 -0700149 addr = ALIGN(addr, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 vma = find_vma(mm, addr);
151 if (TASK_SIZE - len >= addr &&
152 (!vma || addr + len <= vma->vm_start))
153 return addr;
154 }
155
156 start_addr = mm->free_area_cache;
157
Wolfgang Wander1363c3c2005-06-21 17:14:49 -0700158 if (len <= mm->cached_hole_size)
159 start_addr = TASK_UNMAPPED_BASE;
160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161full_search:
Andi Kleena5516432008-07-23 21:27:41 -0700162 addr = ALIGN(start_addr, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
165 /* At this point: (!vma || addr < vma->vm_end). */
166 if (TASK_SIZE - len < addr) {
167 /*
168 * Start a new search - just in case we missed
169 * some holes.
170 */
171 if (start_addr != TASK_UNMAPPED_BASE) {
172 start_addr = TASK_UNMAPPED_BASE;
173 goto full_search;
174 }
175 return -ENOMEM;
176 }
177
178 if (!vma || addr + len <= vma->vm_start)
179 return addr;
Andi Kleena5516432008-07-23 21:27:41 -0700180 addr = ALIGN(vma->vm_end, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 }
182}
183#endif
184
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700185static int
186hugetlbfs_read_actor(struct page *page, unsigned long offset,
187 char __user *buf, unsigned long count,
188 unsigned long size)
189{
190 char *kaddr;
191 unsigned long left, copied = 0;
192 int i, chunksize;
193
194 if (size > count)
195 size = count;
196
197 /* Find which 4k chunk and offset with in that chunk */
198 i = offset >> PAGE_CACHE_SHIFT;
199 offset = offset & ~PAGE_CACHE_MASK;
200
201 while (size) {
202 chunksize = PAGE_CACHE_SIZE;
203 if (offset)
204 chunksize -= offset;
205 if (chunksize > size)
206 chunksize = size;
207 kaddr = kmap(&page[i]);
208 left = __copy_to_user(buf, kaddr + offset, chunksize);
209 kunmap(&page[i]);
210 if (left) {
211 copied += (chunksize - left);
212 break;
213 }
214 offset = 0;
215 size -= chunksize;
216 buf += chunksize;
217 copied += chunksize;
218 i++;
219 }
220 return copied ? copied : -EFAULT;
221}
222
223/*
224 * Support for read() - Find the page attached to f_mapping and copy out the
225 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
226 * since it has PAGE_CACHE_SIZE assumptions.
227 */
228static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
229 size_t len, loff_t *ppos)
230{
Andi Kleena5516432008-07-23 21:27:41 -0700231 struct hstate *h = hstate_file(filp);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700232 struct address_space *mapping = filp->f_mapping;
233 struct inode *inode = mapping->host;
Andi Kleena5516432008-07-23 21:27:41 -0700234 unsigned long index = *ppos >> huge_page_shift(h);
235 unsigned long offset = *ppos & ~huge_page_mask(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700236 unsigned long end_index;
237 loff_t isize;
238 ssize_t retval = 0;
239
240 mutex_lock(&inode->i_mutex);
241
242 /* validate length */
243 if (len == 0)
244 goto out;
245
246 isize = i_size_read(inode);
247 if (!isize)
248 goto out;
249
Andi Kleena5516432008-07-23 21:27:41 -0700250 end_index = (isize - 1) >> huge_page_shift(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700251 for (;;) {
252 struct page *page;
Andi Kleena5516432008-07-23 21:27:41 -0700253 unsigned long nr, ret;
Roel Kluin91bf189c2009-01-06 14:40:14 -0800254 int ra;
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))
Roel Kluin91bf189c2009-01-06 14:40:14 -0800277 ra = -EFAULT;
278 else
279 ra = 0;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700280 } else {
281 /*
282 * We have the page, copy it to user space buffer.
283 */
Roel Kluin91bf189c2009-01-06 14:40:14 -0800284 ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
285 ret = ra;
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 if (page)
291 page_cache_release(page);
292 goto out;
293 }
294
295 offset += ret;
296 retval += ret;
297 len -= ret;
Andi Kleena5516432008-07-23 21:27:41 -0700298 index += offset >> huge_page_shift(h);
299 offset &= ~huge_page_mask(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700300
301 if (page)
302 page_cache_release(page);
303
304 /* short read or no more work */
305 if ((ret != nr) || (len == 0))
306 break;
307 }
308out:
Andi Kleena5516432008-07-23 21:27:41 -0700309 *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700310 mutex_unlock(&inode->i_mutex);
311 return retval;
312}
313
Nick Piggin800d15a2007-10-16 01:25:03 -0700314static int hugetlbfs_write_begin(struct file *file,
315 struct address_space *mapping,
316 loff_t pos, unsigned len, unsigned flags,
317 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
319 return -EINVAL;
320}
321
Nick Piggin800d15a2007-10-16 01:25:03 -0700322static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
323 loff_t pos, unsigned len, unsigned copied,
324 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
Nick Piggin800d15a2007-10-16 01:25:03 -0700326 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 return -EINVAL;
328}
329
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330static void truncate_huge_page(struct page *page)
331{
Linus Torvaldsfba25912006-12-20 13:46:42 -0800332 cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 ClearPageUptodate(page);
334 remove_from_page_cache(page);
335 put_page(page);
336}
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
374static void hugetlbfs_delete_inode(struct inode *inode)
375{
David Gibsonb45b5bd2006-03-22 00:08:55 -0800376 truncate_hugepages(inode, 0);
Christoph Hellwig149f4212005-10-29 18:16:43 -0700377 clear_inode(inode);
378}
379
Josh Triplettddc0a512006-09-29 01:59:27 -0700380static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
Jan Kara22fe40422009-09-18 13:05:44 -0700382 if (generic_detach_inode(inode)) {
383 truncate_hugepages(inode, 0);
384 clear_inode(inode);
385 destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387}
388
389static void hugetlbfs_drop_inode(struct inode *inode)
390{
391 if (!inode->i_nlink)
Christoph Hellwig6b09b9d2005-10-29 18:16:44 -0700392 generic_delete_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 else
394 hugetlbfs_forget_inode(inode);
395}
396
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397static inline void
Hugh Dickins856fc292006-10-28 10:38:43 -0700398hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
400 struct vm_area_struct *vma;
401 struct prio_tree_iter iter;
402
Hugh Dickins856fc292006-10-28 10:38:43 -0700403 vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 unsigned long v_offset;
405
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 /*
Hugh Dickins856fc292006-10-28 10:38:43 -0700407 * Can the expression below overflow on 32-bit arches?
408 * No, because the prio_tree returns us only those vmas
409 * which overlap the truncated area starting at pgoff,
410 * and no vma on a 32-bit arch can span beyond the 4GB.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 */
Hugh Dickins856fc292006-10-28 10:38:43 -0700412 if (vma->vm_pgoff < pgoff)
413 v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
414 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 v_offset = 0;
416
Chen, Kenneth W502717f2006-10-11 01:20:46 -0700417 __unmap_hugepage_range(vma,
Mel Gorman04f2cbe2008-07-23 21:27:25 -0700418 vma->vm_start + v_offset, vma->vm_end, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 }
420}
421
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
423{
Hugh Dickins856fc292006-10-28 10:38:43 -0700424 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 struct address_space *mapping = inode->i_mapping;
Andi Kleena5516432008-07-23 21:27:41 -0700426 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
Andi Kleena5516432008-07-23 21:27:41 -0700428 BUG_ON(offset & ~huge_page_mask(h));
Hugh Dickins856fc292006-10-28 10:38:43 -0700429 pgoff = offset >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
Ken Chen7aa91e12007-10-16 01:26:21 -0700431 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 spin_lock(&mapping->i_mmap_lock);
433 if (!prio_tree_empty(&mapping->i_mmap))
434 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
435 spin_unlock(&mapping->i_mmap_lock);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800436 truncate_hugepages(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 return 0;
438}
439
440static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
441{
442 struct inode *inode = dentry->d_inode;
Andi Kleena5516432008-07-23 21:27:41 -0700443 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 int error;
445 unsigned int ia_valid = attr->ia_valid;
446
447 BUG_ON(!inode);
448
449 error = inode_change_ok(inode, attr);
450 if (error)
451 goto out;
452
453 if (ia_valid & ATTR_SIZE) {
454 error = -EINVAL;
Andi Kleena5516432008-07-23 21:27:41 -0700455 if (!(attr->ia_size & ~huge_page_mask(h)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 error = hugetlb_vmtruncate(inode, attr->ia_size);
457 if (error)
458 goto out;
459 attr->ia_valid &= ~ATTR_SIZE;
460 }
461 error = inode_setattr(inode, attr);
462out:
463 return error;
464}
465
466static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
467 gid_t gid, int mode, dev_t dev)
468{
469 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
471 inode = new_inode(sb);
472 if (inode) {
473 struct hugetlbfs_inode_info *info;
474 inode->i_mode = mode;
475 inode->i_uid = uid;
476 inode->i_gid = gid;
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 }
509 }
510 return inode;
511}
512
513/*
514 * File creation. Allocate an inode, and we're done..
515 */
516static int hugetlbfs_mknod(struct inode *dir,
517 struct dentry *dentry, int mode, dev_t dev)
518{
519 struct inode *inode;
520 int error = -ENOSPC;
521 gid_t gid;
522
523 if (dir->i_mode & S_ISGID) {
524 gid = dir->i_gid;
525 if (S_ISDIR(mode))
526 mode |= S_ISGID;
527 } else {
David Howells77c70de2008-11-14 10:38:56 +1100528 gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 }
David Howells77c70de2008-11-14 10:38:56 +1100530 inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 if (inode) {
532 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
533 d_instantiate(dentry, inode);
534 dget(dentry); /* Extra count - pin the dentry in core */
535 error = 0;
536 }
537 return error;
538}
539
540static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
541{
542 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
543 if (!retval)
Dave Hansend8c76e62006-09-30 23:29:04 -0700544 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 return retval;
546}
547
548static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
549{
550 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
551}
552
553static int hugetlbfs_symlink(struct inode *dir,
554 struct dentry *dentry, const char *symname)
555{
556 struct inode *inode;
557 int error = -ENOSPC;
558 gid_t gid;
559
560 if (dir->i_mode & S_ISGID)
561 gid = dir->i_gid;
562 else
David Howells77c70de2008-11-14 10:38:56 +1100563 gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
David Howells77c70de2008-11-14 10:38:56 +1100565 inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 gid, S_IFLNK|S_IRWXUGO, 0);
567 if (inode) {
568 int l = strlen(symname)+1;
569 error = page_symlink(inode, symname, l);
570 if (!error) {
571 d_instantiate(dentry, inode);
572 dget(dentry);
573 } else
574 iput(inode);
575 }
576 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
577
578 return error;
579}
580
581/*
Ken Chen6649a382007-02-08 14:20:27 -0800582 * mark the head page dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 */
584static int hugetlbfs_set_page_dirty(struct page *page)
585{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700586 struct page *head = compound_head(page);
Ken Chen6649a382007-02-08 14:20:27 -0800587
588 SetPageDirty(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 return 0;
590}
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() */
603 if (sbinfo->max_blocks >= 0) {
604 buf->f_blocks = sbinfo->max_blocks;
605 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
606 buf->f_files = sbinfo->max_inodes;
607 buf->f_ffree = sbinfo->free_inodes;
608 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 spin_unlock(&sbinfo->stat_lock);
610 }
611 buf->f_namelen = NAME_MAX;
612 return 0;
613}
614
615static void hugetlbfs_put_super(struct super_block *sb)
616{
617 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
618
619 if (sbi) {
620 sb->s_fs_info = NULL;
621 kfree(sbi);
622 }
623}
624
Christoph Hellwig96527982005-10-29 18:16:42 -0700625static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
626{
627 if (sbinfo->free_inodes >= 0) {
628 spin_lock(&sbinfo->stat_lock);
629 if (unlikely(!sbinfo->free_inodes)) {
630 spin_unlock(&sbinfo->stat_lock);
631 return 0;
632 }
633 sbinfo->free_inodes--;
634 spin_unlock(&sbinfo->stat_lock);
635 }
636
637 return 1;
638}
639
640static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
641{
642 if (sbinfo->free_inodes >= 0) {
643 spin_lock(&sbinfo->stat_lock);
644 sbinfo->free_inodes++;
645 spin_unlock(&sbinfo->stat_lock);
646 }
647}
648
649
Christoph Lametere18b8902006-12-06 20:33:20 -0800650static struct kmem_cache *hugetlbfs_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
652static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
653{
Christoph Hellwig96527982005-10-29 18:16:42 -0700654 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 struct hugetlbfs_inode_info *p;
656
Christoph Hellwig96527982005-10-29 18:16:42 -0700657 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 return NULL;
Christoph Lametere94b1762006-12-06 20:33:17 -0800659 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
Christoph Hellwig96527982005-10-29 18:16:42 -0700660 if (unlikely(!p)) {
661 hugetlbfs_inc_free_inodes(sbinfo);
662 return NULL;
663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 return &p->vfs_inode;
665}
666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667static void hugetlbfs_destroy_inode(struct inode *inode)
668{
Christoph Hellwig96527982005-10-29 18:16:42 -0700669 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
671 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
672}
673
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700674static const struct address_space_operations hugetlbfs_aops = {
Nick Piggin800d15a2007-10-16 01:25:03 -0700675 .write_begin = hugetlbfs_write_begin,
676 .write_end = hugetlbfs_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 .set_page_dirty = hugetlbfs_set_page_dirty,
678};
679
Christoph Hellwig96527982005-10-29 18:16:42 -0700680
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700681static void init_once(void *foo)
Christoph Hellwig96527982005-10-29 18:16:42 -0700682{
683 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
684
Christoph Lametera35afb82007-05-16 22:10:57 -0700685 inode_init_once(&ei->vfs_inode);
Christoph Hellwig96527982005-10-29 18:16:42 -0700686}
687
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800688const struct file_operations hugetlbfs_file_operations = {
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700689 .read = hugetlbfs_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 .mmap = hugetlbfs_file_mmap,
Christoph Hellwig1b061d92010-05-26 17:53:41 +0200691 .fsync = noop_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 .get_unmapped_area = hugetlb_get_unmapped_area,
693};
694
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800695static const struct inode_operations hugetlbfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 .create = hugetlbfs_create,
697 .lookup = simple_lookup,
698 .link = simple_link,
699 .unlink = simple_unlink,
700 .symlink = hugetlbfs_symlink,
701 .mkdir = hugetlbfs_mkdir,
702 .rmdir = simple_rmdir,
703 .mknod = hugetlbfs_mknod,
704 .rename = simple_rename,
705 .setattr = hugetlbfs_setattr,
706};
707
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800708static const struct inode_operations hugetlbfs_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 .setattr = hugetlbfs_setattr,
710};
711
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800712static const struct super_operations hugetlbfs_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 .alloc_inode = hugetlbfs_alloc_inode,
714 .destroy_inode = hugetlbfs_destroy_inode,
715 .statfs = hugetlbfs_statfs,
Christoph Hellwig149f4212005-10-29 18:16:43 -0700716 .delete_inode = hugetlbfs_delete_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 .drop_inode = hugetlbfs_drop_inode,
718 .put_super = hugetlbfs_put_super,
Miklos Szeredi10f19a82008-02-08 04:21:45 -0800719 .show_options = generic_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720};
721
722static int
723hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
724{
Randy Dunlape73a75f2007-07-15 23:40:52 -0700725 char *p, *rest;
726 substring_t args[MAX_OPT_ARGS];
727 int option;
Andi Kleena137e1c2008-07-23 21:27:43 -0700728 unsigned long long size = 0;
729 enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
731 if (!options)
732 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
Randy Dunlape73a75f2007-07-15 23:40:52 -0700734 while ((p = strsep(&options, ",")) != NULL) {
735 int token;
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -0700736 if (!*p)
737 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
Randy Dunlape73a75f2007-07-15 23:40:52 -0700739 token = match_token(p, tokens, args);
740 switch (token) {
741 case Opt_uid:
742 if (match_int(&args[0], &option))
743 goto bad_val;
744 pconfig->uid = option;
745 break;
746
747 case Opt_gid:
748 if (match_int(&args[0], &option))
749 goto bad_val;
750 pconfig->gid = option;
751 break;
752
753 case Opt_mode:
754 if (match_octal(&args[0], &option))
755 goto bad_val;
Ken Chen75897d62008-02-04 22:28:36 -0800756 pconfig->mode = option & 01777U;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700757 break;
758
759 case Opt_size: {
Randy Dunlape73a75f2007-07-15 23:40:52 -0700760 /* memparse() will accept a K/M/G without a digit */
761 if (!isdigit(*args[0].from))
762 goto bad_val;
763 size = memparse(args[0].from, &rest);
Andi Kleena137e1c2008-07-23 21:27:43 -0700764 setsize = SIZE_STD;
765 if (*rest == '%')
766 setsize = SIZE_PERCENT;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700767 break;
768 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Randy Dunlape73a75f2007-07-15 23:40:52 -0700770 case Opt_nr_inodes:
771 /* memparse() will accept a K/M/G without a digit */
772 if (!isdigit(*args[0].from))
773 goto bad_val;
774 pconfig->nr_inodes = memparse(args[0].from, &rest);
775 break;
776
Andi Kleena137e1c2008-07-23 21:27:43 -0700777 case Opt_pagesize: {
778 unsigned long ps;
779 ps = memparse(args[0].from, &rest);
780 pconfig->hstate = size_to_hstate(ps);
781 if (!pconfig->hstate) {
782 printk(KERN_ERR
783 "hugetlbfs: Unsupported page size %lu MB\n",
784 ps >> 20);
785 return -EINVAL;
786 }
787 break;
788 }
789
Randy Dunlape73a75f2007-07-15 23:40:52 -0700790 default:
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -0700791 printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
792 p);
793 return -EINVAL;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700794 break;
795 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 }
Andi Kleena137e1c2008-07-23 21:27:43 -0700797
798 /* Do size after hstate is set up */
799 if (setsize > NO_SIZE) {
800 struct hstate *h = pconfig->hstate;
801 if (setsize == SIZE_PERCENT) {
802 size <<= huge_page_shift(h);
803 size *= h->max_huge_pages;
804 do_div(size, 100);
805 }
806 pconfig->nr_blocks = (size >> huge_page_shift(h));
807 }
808
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 return 0;
Randy Dunlape73a75f2007-07-15 23:40:52 -0700810
811bad_val:
812 printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
813 args[0].from, p);
Akinobu Mitac12ddba2009-04-21 12:24:05 -0700814 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815}
816
817static int
818hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
819{
820 struct inode * inode;
821 struct dentry * root;
822 int ret;
823 struct hugetlbfs_config config;
824 struct hugetlbfs_sb_info *sbinfo;
825
Miklos Szeredi10f19a82008-02-08 04:21:45 -0800826 save_mount_options(sb, data);
827
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 config.nr_blocks = -1; /* No limit on size by default */
829 config.nr_inodes = -1; /* No limit on number of inodes by default */
David Howells77c70de2008-11-14 10:38:56 +1100830 config.uid = current_fsuid();
831 config.gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 config.mode = 0755;
Andi Kleena137e1c2008-07-23 21:27:43 -0700833 config.hstate = &default_hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 ret = hugetlbfs_parse_options(data, &config);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 if (ret)
836 return ret;
837
838 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
839 if (!sbinfo)
840 return -ENOMEM;
841 sb->s_fs_info = sbinfo;
Andi Kleena137e1c2008-07-23 21:27:43 -0700842 sbinfo->hstate = config.hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 spin_lock_init(&sbinfo->stat_lock);
844 sbinfo->max_blocks = config.nr_blocks;
845 sbinfo->free_blocks = config.nr_blocks;
846 sbinfo->max_inodes = config.nr_inodes;
847 sbinfo->free_inodes = config.nr_inodes;
848 sb->s_maxbytes = MAX_LFS_FILESIZE;
Andi Kleena137e1c2008-07-23 21:27:43 -0700849 sb->s_blocksize = huge_page_size(config.hstate);
850 sb->s_blocksize_bits = huge_page_shift(config.hstate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 sb->s_magic = HUGETLBFS_MAGIC;
852 sb->s_op = &hugetlbfs_ops;
853 sb->s_time_gran = 1;
854 inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
855 S_IFDIR | config.mode, 0);
856 if (!inode)
857 goto out_free;
858
859 root = d_alloc_root(inode);
860 if (!root) {
861 iput(inode);
862 goto out_free;
863 }
864 sb->s_root = root;
865 return 0;
866out_free:
867 kfree(sbinfo);
868 return -ENOMEM;
869}
870
Adam Litke9a119c02007-11-14 16:59:41 -0800871int hugetlb_get_quota(struct address_space *mapping, long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872{
873 int ret = 0;
874 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
875
876 if (sbinfo->free_blocks > -1) {
877 spin_lock(&sbinfo->stat_lock);
Adam Litke9a119c02007-11-14 16:59:41 -0800878 if (sbinfo->free_blocks - delta >= 0)
879 sbinfo->free_blocks -= delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 else
881 ret = -ENOMEM;
882 spin_unlock(&sbinfo->stat_lock);
883 }
884
885 return ret;
886}
887
Adam Litke9a119c02007-11-14 16:59:41 -0800888void hugetlb_put_quota(struct address_space *mapping, long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889{
890 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
891
892 if (sbinfo->free_blocks > -1) {
893 spin_lock(&sbinfo->stat_lock);
Adam Litke9a119c02007-11-14 16:59:41 -0800894 sbinfo->free_blocks += delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 spin_unlock(&sbinfo->stat_lock);
896 }
897}
898
David Howells454e2392006-06-23 02:02:57 -0700899static int hugetlbfs_get_sb(struct file_system_type *fs_type,
900 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
David Howells454e2392006-06-23 02:02:57 -0700902 return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903}
904
905static struct file_system_type hugetlbfs_fs_type = {
906 .name = "hugetlbfs",
907 .get_sb = hugetlbfs_get_sb,
908 .kill_sb = kill_litter_super,
909};
910
911static struct vfsmount *hugetlbfs_vfsmount;
912
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700913static int can_do_hugetlb_shm(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914{
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700915 return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916}
917
Hugh Dickins353d5c32009-08-24 16:30:28 +0100918struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag,
Eric B Munson6bfde052009-09-21 17:03:43 -0700919 struct user_struct **user, int creat_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920{
921 int error = -ENOMEM;
922 struct file *file;
923 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400924 struct path path;
925 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 struct qstr quick_string;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Hugh Dickins353d5c32009-08-24 16:30:28 +0100928 *user = NULL;
Akinobu Mita5bc98592007-05-06 14:50:18 -0700929 if (!hugetlbfs_vfsmount)
930 return ERR_PTR(-ENOENT);
931
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -0700932 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
Hugh Dickins353d5c32009-08-24 16:30:28 +0100933 *user = current_user();
934 if (user_shm_lock(size, *user)) {
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -0700935 WARN_ONCE(1,
936 "Using mlock ulimits for SHM_HUGETLB deprecated\n");
Hugh Dickins353d5c32009-08-24 16:30:28 +0100937 } else {
938 *user = NULL;
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -0700939 return ERR_PTR(-EPERM);
Hugh Dickins353d5c32009-08-24 16:30:28 +0100940 }
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -0700941 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 root = hugetlbfs_vfsmount->mnt_root;
Eric W. Biederman9d665862007-06-16 10:16:16 -0700944 quick_string.name = name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 quick_string.len = strlen(quick_string.name);
946 quick_string.hash = 0;
Al Viro2c48b9c2009-08-09 00:52:35 +0400947 path.dentry = d_alloc(root, &quick_string);
948 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 goto out_shm_unlock;
950
Al Viro2c48b9c2009-08-09 00:52:35 +0400951 path.mnt = mntget(hugetlbfs_vfsmount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 error = -ENOSPC;
David Howells77c70de2008-11-14 10:38:56 +1100953 inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
954 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 if (!inode)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700956 goto out_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
David Gibsonb45b5bd2006-03-22 00:08:55 -0800958 error = -ENOMEM;
Andi Kleena5516432008-07-23 21:27:41 -0700959 if (hugetlb_reserve_pages(inode, 0,
Mel Gorman5a6fe122009-02-10 14:02:27 +0000960 size >> huge_page_shift(hstate_inode(inode)), NULL,
961 acctflag))
David Gibsonb45b5bd2006-03-22 00:08:55 -0800962 goto out_inode;
963
Al Viro2c48b9c2009-08-09 00:52:35 +0400964 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 inode->i_size = size;
966 inode->i_nlink = 0;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700967
968 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +0400969 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Dave Hansence8d2cd2007-10-16 23:31:13 -0700970 &hugetlbfs_file_operations);
971 if (!file)
Al Virob4d232e2008-02-23 05:59:19 -0500972 goto out_dentry; /* inode is already attached */
Dave Hansence8d2cd2007-10-16 23:31:13 -0700973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 return file;
975
David Gibsonb45b5bd2006-03-22 00:08:55 -0800976out_inode:
977 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978out_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +0400979 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980out_shm_unlock:
Hugh Dickins353d5c32009-08-24 16:30:28 +0100981 if (*user) {
982 user_shm_unlock(size, *user);
983 *user = NULL;
984 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 return ERR_PTR(error);
986}
987
988static int __init init_hugetlbfs_fs(void)
989{
990 int error;
991 struct vfsmount *vfsmount;
992
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700993 error = bdi_init(&hugetlbfs_backing_dev_info);
994 if (error)
995 return error;
996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
998 sizeof(struct hugetlbfs_inode_info),
Paul Mundt20c2df82007-07-20 10:11:58 +0900999 0, 0, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 if (hugetlbfs_inode_cachep == NULL)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001001 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
1003 error = register_filesystem(&hugetlbfs_fs_type);
1004 if (error)
1005 goto out;
1006
1007 vfsmount = kern_mount(&hugetlbfs_fs_type);
1008
1009 if (!IS_ERR(vfsmount)) {
1010 hugetlbfs_vfsmount = vfsmount;
1011 return 0;
1012 }
1013
1014 error = PTR_ERR(vfsmount);
1015
1016 out:
1017 if (error)
1018 kmem_cache_destroy(hugetlbfs_inode_cachep);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001019 out2:
1020 bdi_destroy(&hugetlbfs_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 return error;
1022}
1023
1024static void __exit exit_hugetlbfs_fs(void)
1025{
1026 kmem_cache_destroy(hugetlbfs_inode_cachep);
1027 unregister_filesystem(&hugetlbfs_fs_type);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001028 bdi_destroy(&hugetlbfs_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029}
1030
1031module_init(init_hugetlbfs_fs)
1032module_exit(exit_hugetlbfs_fs)
1033
1034MODULE_LICENSE("GPL");