hugepage: fix broken check for offset alignment in hugepage mappings
[linux-2.6.git] / fs / hugetlbfs / inode.c
1 /*
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>
16 #include <linux/kernel.h>
17 #include <linux/writeback.h>
18 #include <linux/pagemap.h>
19 #include <linux/highmem.h>
20 #include <linux/init.h>
21 #include <linux/string.h>
22 #include <linux/capability.h>
23 #include <linux/ctype.h>
24 #include <linux/backing-dev.h>
25 #include <linux/hugetlb.h>
26 #include <linux/pagevec.h>
27 #include <linux/parser.h>
28 #include <linux/mman.h>
29 #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
40 static const struct super_operations hugetlbfs_ops;
41 static const struct address_space_operations hugetlbfs_aops;
42 const struct file_operations hugetlbfs_file_operations;
43 static const struct inode_operations hugetlbfs_dir_inode_operations;
44 static const struct inode_operations hugetlbfs_inode_operations;
45
46 static struct backing_dev_info hugetlbfs_backing_dev_info = {
47         .ra_pages       = 0,    /* No readahead */
48         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
49 };
50
51 int sysctl_hugetlb_shm_group;
52
53 enum {
54         Opt_size, Opt_nr_inodes,
55         Opt_mode, Opt_uid, Opt_gid,
56         Opt_err,
57 };
58
59 static match_table_t tokens = {
60         {Opt_size,      "size=%s"},
61         {Opt_nr_inodes, "nr_inodes=%s"},
62         {Opt_mode,      "mode=%o"},
63         {Opt_uid,       "uid=%u"},
64         {Opt_gid,       "gid=%u"},
65         {Opt_err,       NULL},
66 };
67
68 static 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
78 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
79 {
80         struct inode *inode = file->f_path.dentry->d_inode;
81         loff_t len, vma_len;
82         int ret;
83
84         /*
85          * vma address alignment (but not the pgoff alignment) has
86          * already been checked by prepare_hugepage_range.  If you add
87          * any error returns here, do so after setting VM_HUGETLB, so
88          * is_vm_hugetlb_page tests below unmap_region go the right
89          * way when do_mmap_pgoff unwinds (may be important on powerpc
90          * and ia64).
91          */
92         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
93         vma->vm_ops = &hugetlb_vm_ops;
94
95         if (vma->vm_pgoff & ~(HPAGE_MASK >> PAGE_SHIFT))
96                 return -EINVAL;
97
98         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
99
100         mutex_lock(&inode->i_mutex);
101         file_accessed(file);
102
103         ret = -ENOMEM;
104         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
105
106         if (vma->vm_flags & VM_MAYSHARE &&
107             hugetlb_reserve_pages(inode, vma->vm_pgoff >> (HPAGE_SHIFT-PAGE_SHIFT),
108                                   len >> HPAGE_SHIFT))
109                 goto out;
110
111         ret = 0;
112         hugetlb_prefault_arch_hook(vma->vm_mm);
113         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
114                 inode->i_size = len;
115 out:
116         mutex_unlock(&inode->i_mutex);
117
118         return ret;
119 }
120
121 /*
122  * Called under down_write(mmap_sem).
123  */
124
125 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
126 static unsigned long
127 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
128                 unsigned long len, unsigned long pgoff, unsigned long flags)
129 {
130         struct mm_struct *mm = current->mm;
131         struct vm_area_struct *vma;
132         unsigned long start_addr;
133
134         if (len & ~HPAGE_MASK)
135                 return -EINVAL;
136         if (len > TASK_SIZE)
137                 return -ENOMEM;
138
139         if (flags & MAP_FIXED) {
140                 if (prepare_hugepage_range(addr, len))
141                         return -EINVAL;
142                 return addr;
143         }
144
145         if (addr) {
146                 addr = ALIGN(addr, HPAGE_SIZE);
147                 vma = find_vma(mm, addr);
148                 if (TASK_SIZE - len >= addr &&
149                     (!vma || addr + len <= vma->vm_start))
150                         return addr;
151         }
152
153         start_addr = mm->free_area_cache;
154
155         if (len <= mm->cached_hole_size)
156                 start_addr = TASK_UNMAPPED_BASE;
157
158 full_search:
159         addr = ALIGN(start_addr, HPAGE_SIZE);
160
161         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
162                 /* At this point:  (!vma || addr < vma->vm_end). */
163                 if (TASK_SIZE - len < addr) {
164                         /*
165                          * Start a new search - just in case we missed
166                          * some holes.
167                          */
168                         if (start_addr != TASK_UNMAPPED_BASE) {
169                                 start_addr = TASK_UNMAPPED_BASE;
170                                 goto full_search;
171                         }
172                         return -ENOMEM;
173                 }
174
175                 if (!vma || addr + len <= vma->vm_start)
176                         return addr;
177                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
178         }
179 }
180 #endif
181
182 /*
183  * Read a page. Again trivial. If it didn't already exist
184  * in the page cache, it is zero-filled.
185  */
186 static int hugetlbfs_readpage(struct file *file, struct page * page)
187 {
188         unlock_page(page);
189         return -EINVAL;
190 }
191
192 static int hugetlbfs_prepare_write(struct file *file,
193                         struct page *page, unsigned offset, unsigned to)
194 {
195         return -EINVAL;
196 }
197
198 static int hugetlbfs_commit_write(struct file *file,
199                         struct page *page, unsigned offset, unsigned to)
200 {
201         return -EINVAL;
202 }
203
204 static void truncate_huge_page(struct page *page)
205 {
206         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
207         ClearPageUptodate(page);
208         remove_from_page_cache(page);
209         put_page(page);
210 }
211
212 static void truncate_hugepages(struct inode *inode, loff_t lstart)
213 {
214         struct address_space *mapping = &inode->i_data;
215         const pgoff_t start = lstart >> HPAGE_SHIFT;
216         struct pagevec pvec;
217         pgoff_t next;
218         int i, freed = 0;
219
220         pagevec_init(&pvec, 0);
221         next = start;
222         while (1) {
223                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
224                         if (next == start)
225                                 break;
226                         next = start;
227                         continue;
228                 }
229
230                 for (i = 0; i < pagevec_count(&pvec); ++i) {
231                         struct page *page = pvec.pages[i];
232
233                         lock_page(page);
234                         if (page->index > next)
235                                 next = page->index;
236                         ++next;
237                         truncate_huge_page(page);
238                         unlock_page(page);
239                         hugetlb_put_quota(mapping);
240                         freed++;
241                 }
242                 huge_pagevec_release(&pvec);
243         }
244         BUG_ON(!lstart && mapping->nrpages);
245         hugetlb_unreserve_pages(inode, start, freed);
246 }
247
248 static void hugetlbfs_delete_inode(struct inode *inode)
249 {
250         truncate_hugepages(inode, 0);
251         clear_inode(inode);
252 }
253
254 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
255 {
256         struct super_block *sb = inode->i_sb;
257
258         if (!hlist_unhashed(&inode->i_hash)) {
259                 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
260                         list_move(&inode->i_list, &inode_unused);
261                 inodes_stat.nr_unused++;
262                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
263                         spin_unlock(&inode_lock);
264                         return;
265                 }
266                 inode->i_state |= I_WILL_FREE;
267                 spin_unlock(&inode_lock);
268                 /*
269                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
270                  * in our backing_dev_info.
271                  */
272                 write_inode_now(inode, 1);
273                 spin_lock(&inode_lock);
274                 inode->i_state &= ~I_WILL_FREE;
275                 inodes_stat.nr_unused--;
276                 hlist_del_init(&inode->i_hash);
277         }
278         list_del_init(&inode->i_list);
279         list_del_init(&inode->i_sb_list);
280         inode->i_state |= I_FREEING;
281         inodes_stat.nr_inodes--;
282         spin_unlock(&inode_lock);
283         truncate_hugepages(inode, 0);
284         clear_inode(inode);
285         destroy_inode(inode);
286 }
287
288 static void hugetlbfs_drop_inode(struct inode *inode)
289 {
290         if (!inode->i_nlink)
291                 generic_delete_inode(inode);
292         else
293                 hugetlbfs_forget_inode(inode);
294 }
295
296 static inline void
297 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
298 {
299         struct vm_area_struct *vma;
300         struct prio_tree_iter iter;
301
302         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
303                 unsigned long v_offset;
304
305                 /*
306                  * Can the expression below overflow on 32-bit arches?
307                  * No, because the prio_tree returns us only those vmas
308                  * which overlap the truncated area starting at pgoff,
309                  * and no vma on a 32-bit arch can span beyond the 4GB.
310                  */
311                 if (vma->vm_pgoff < pgoff)
312                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
313                 else
314                         v_offset = 0;
315
316                 __unmap_hugepage_range(vma,
317                                 vma->vm_start + v_offset, vma->vm_end);
318         }
319 }
320
321 /*
322  * Expanding truncates are not allowed.
323  */
324 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
325 {
326         pgoff_t pgoff;
327         struct address_space *mapping = inode->i_mapping;
328
329         if (offset > inode->i_size)
330                 return -EINVAL;
331
332         BUG_ON(offset & ~HPAGE_MASK);
333         pgoff = offset >> PAGE_SHIFT;
334
335         inode->i_size = offset;
336         spin_lock(&mapping->i_mmap_lock);
337         if (!prio_tree_empty(&mapping->i_mmap))
338                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
339         spin_unlock(&mapping->i_mmap_lock);
340         truncate_hugepages(inode, offset);
341         return 0;
342 }
343
344 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
345 {
346         struct inode *inode = dentry->d_inode;
347         int error;
348         unsigned int ia_valid = attr->ia_valid;
349
350         BUG_ON(!inode);
351
352         error = inode_change_ok(inode, attr);
353         if (error)
354                 goto out;
355
356         if (ia_valid & ATTR_SIZE) {
357                 error = -EINVAL;
358                 if (!(attr->ia_size & ~HPAGE_MASK))
359                         error = hugetlb_vmtruncate(inode, attr->ia_size);
360                 if (error)
361                         goto out;
362                 attr->ia_valid &= ~ATTR_SIZE;
363         }
364         error = inode_setattr(inode, attr);
365 out:
366         return error;
367 }
368
369 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid, 
370                                         gid_t gid, int mode, dev_t dev)
371 {
372         struct inode *inode;
373
374         inode = new_inode(sb);
375         if (inode) {
376                 struct hugetlbfs_inode_info *info;
377                 inode->i_mode = mode;
378                 inode->i_uid = uid;
379                 inode->i_gid = gid;
380                 inode->i_blocks = 0;
381                 inode->i_mapping->a_ops = &hugetlbfs_aops;
382                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
383                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
384                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
385                 info = HUGETLBFS_I(inode);
386                 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
387                 switch (mode & S_IFMT) {
388                 default:
389                         init_special_inode(inode, mode, dev);
390                         break;
391                 case S_IFREG:
392                         inode->i_op = &hugetlbfs_inode_operations;
393                         inode->i_fop = &hugetlbfs_file_operations;
394                         break;
395                 case S_IFDIR:
396                         inode->i_op = &hugetlbfs_dir_inode_operations;
397                         inode->i_fop = &simple_dir_operations;
398
399                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
400                         inc_nlink(inode);
401                         break;
402                 case S_IFLNK:
403                         inode->i_op = &page_symlink_inode_operations;
404                         break;
405                 }
406         }
407         return inode;
408 }
409
410 /*
411  * File creation. Allocate an inode, and we're done..
412  */
413 static int hugetlbfs_mknod(struct inode *dir,
414                         struct dentry *dentry, int mode, dev_t dev)
415 {
416         struct inode *inode;
417         int error = -ENOSPC;
418         gid_t gid;
419
420         if (dir->i_mode & S_ISGID) {
421                 gid = dir->i_gid;
422                 if (S_ISDIR(mode))
423                         mode |= S_ISGID;
424         } else {
425                 gid = current->fsgid;
426         }
427         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
428         if (inode) {
429                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
430                 d_instantiate(dentry, inode);
431                 dget(dentry);   /* Extra count - pin the dentry in core */
432                 error = 0;
433         }
434         return error;
435 }
436
437 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
438 {
439         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
440         if (!retval)
441                 inc_nlink(dir);
442         return retval;
443 }
444
445 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
446 {
447         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
448 }
449
450 static int hugetlbfs_symlink(struct inode *dir,
451                         struct dentry *dentry, const char *symname)
452 {
453         struct inode *inode;
454         int error = -ENOSPC;
455         gid_t gid;
456
457         if (dir->i_mode & S_ISGID)
458                 gid = dir->i_gid;
459         else
460                 gid = current->fsgid;
461
462         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
463                                         gid, S_IFLNK|S_IRWXUGO, 0);
464         if (inode) {
465                 int l = strlen(symname)+1;
466                 error = page_symlink(inode, symname, l);
467                 if (!error) {
468                         d_instantiate(dentry, inode);
469                         dget(dentry);
470                 } else
471                         iput(inode);
472         }
473         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
474
475         return error;
476 }
477
478 /*
479  * mark the head page dirty
480  */
481 static int hugetlbfs_set_page_dirty(struct page *page)
482 {
483         struct page *head = compound_head(page);
484
485         SetPageDirty(head);
486         return 0;
487 }
488
489 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
490 {
491         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
492
493         buf->f_type = HUGETLBFS_MAGIC;
494         buf->f_bsize = HPAGE_SIZE;
495         if (sbinfo) {
496                 spin_lock(&sbinfo->stat_lock);
497                 /* If no limits set, just report 0 for max/free/used
498                  * blocks, like simple_statfs() */
499                 if (sbinfo->max_blocks >= 0) {
500                         buf->f_blocks = sbinfo->max_blocks;
501                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
502                         buf->f_files = sbinfo->max_inodes;
503                         buf->f_ffree = sbinfo->free_inodes;
504                 }
505                 spin_unlock(&sbinfo->stat_lock);
506         }
507         buf->f_namelen = NAME_MAX;
508         return 0;
509 }
510
511 static void hugetlbfs_put_super(struct super_block *sb)
512 {
513         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
514
515         if (sbi) {
516                 sb->s_fs_info = NULL;
517                 kfree(sbi);
518         }
519 }
520
521 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
522 {
523         if (sbinfo->free_inodes >= 0) {
524                 spin_lock(&sbinfo->stat_lock);
525                 if (unlikely(!sbinfo->free_inodes)) {
526                         spin_unlock(&sbinfo->stat_lock);
527                         return 0;
528                 }
529                 sbinfo->free_inodes--;
530                 spin_unlock(&sbinfo->stat_lock);
531         }
532
533         return 1;
534 }
535
536 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
537 {
538         if (sbinfo->free_inodes >= 0) {
539                 spin_lock(&sbinfo->stat_lock);
540                 sbinfo->free_inodes++;
541                 spin_unlock(&sbinfo->stat_lock);
542         }
543 }
544
545
546 static struct kmem_cache *hugetlbfs_inode_cachep;
547
548 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
549 {
550         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
551         struct hugetlbfs_inode_info *p;
552
553         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
554                 return NULL;
555         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
556         if (unlikely(!p)) {
557                 hugetlbfs_inc_free_inodes(sbinfo);
558                 return NULL;
559         }
560         return &p->vfs_inode;
561 }
562
563 static void hugetlbfs_destroy_inode(struct inode *inode)
564 {
565         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
566         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
567         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
568 }
569
570 static const struct address_space_operations hugetlbfs_aops = {
571         .readpage       = hugetlbfs_readpage,
572         .prepare_write  = hugetlbfs_prepare_write,
573         .commit_write   = hugetlbfs_commit_write,
574         .set_page_dirty = hugetlbfs_set_page_dirty,
575 };
576
577
578 static void init_once(void *foo, struct kmem_cache *cachep, unsigned long flags)
579 {
580         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
581
582         inode_init_once(&ei->vfs_inode);
583 }
584
585 const struct file_operations hugetlbfs_file_operations = {
586         .mmap                   = hugetlbfs_file_mmap,
587         .fsync                  = simple_sync_file,
588         .get_unmapped_area      = hugetlb_get_unmapped_area,
589 };
590
591 static const struct inode_operations hugetlbfs_dir_inode_operations = {
592         .create         = hugetlbfs_create,
593         .lookup         = simple_lookup,
594         .link           = simple_link,
595         .unlink         = simple_unlink,
596         .symlink        = hugetlbfs_symlink,
597         .mkdir          = hugetlbfs_mkdir,
598         .rmdir          = simple_rmdir,
599         .mknod          = hugetlbfs_mknod,
600         .rename         = simple_rename,
601         .setattr        = hugetlbfs_setattr,
602 };
603
604 static const struct inode_operations hugetlbfs_inode_operations = {
605         .setattr        = hugetlbfs_setattr,
606 };
607
608 static const struct super_operations hugetlbfs_ops = {
609         .alloc_inode    = hugetlbfs_alloc_inode,
610         .destroy_inode  = hugetlbfs_destroy_inode,
611         .statfs         = hugetlbfs_statfs,
612         .delete_inode   = hugetlbfs_delete_inode,
613         .drop_inode     = hugetlbfs_drop_inode,
614         .put_super      = hugetlbfs_put_super,
615 };
616
617 static int
618 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
619 {
620         char *p, *rest;
621         substring_t args[MAX_OPT_ARGS];
622         int option;
623
624         if (!options)
625                 return 0;
626
627         while ((p = strsep(&options, ",")) != NULL) {
628                 int token;
629                 if (!*p)
630                         continue;
631
632                 token = match_token(p, tokens, args);
633                 switch (token) {
634                 case Opt_uid:
635                         if (match_int(&args[0], &option))
636                                 goto bad_val;
637                         pconfig->uid = option;
638                         break;
639
640                 case Opt_gid:
641                         if (match_int(&args[0], &option))
642                                 goto bad_val;
643                         pconfig->gid = option;
644                         break;
645
646                 case Opt_mode:
647                         if (match_octal(&args[0], &option))
648                                 goto bad_val;
649                         pconfig->mode = option & 0777U;
650                         break;
651
652                 case Opt_size: {
653                         unsigned long long size;
654                         /* memparse() will accept a K/M/G without a digit */
655                         if (!isdigit(*args[0].from))
656                                 goto bad_val;
657                         size = memparse(args[0].from, &rest);
658                         if (*rest == '%') {
659                                 size <<= HPAGE_SHIFT;
660                                 size *= max_huge_pages;
661                                 do_div(size, 100);
662                         }
663                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
664                         break;
665                 }
666
667                 case Opt_nr_inodes:
668                         /* memparse() will accept a K/M/G without a digit */
669                         if (!isdigit(*args[0].from))
670                                 goto bad_val;
671                         pconfig->nr_inodes = memparse(args[0].from, &rest);
672                         break;
673
674                 default:
675                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
676                                  p);
677                         return -EINVAL;
678                         break;
679                 }
680         }
681         return 0;
682
683 bad_val:
684         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
685                args[0].from, p);
686         return 1;
687 }
688
689 static int
690 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
691 {
692         struct inode * inode;
693         struct dentry * root;
694         int ret;
695         struct hugetlbfs_config config;
696         struct hugetlbfs_sb_info *sbinfo;
697
698         config.nr_blocks = -1; /* No limit on size by default */
699         config.nr_inodes = -1; /* No limit on number of inodes by default */
700         config.uid = current->fsuid;
701         config.gid = current->fsgid;
702         config.mode = 0755;
703         ret = hugetlbfs_parse_options(data, &config);
704         if (ret)
705                 return ret;
706
707         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
708         if (!sbinfo)
709                 return -ENOMEM;
710         sb->s_fs_info = sbinfo;
711         spin_lock_init(&sbinfo->stat_lock);
712         sbinfo->max_blocks = config.nr_blocks;
713         sbinfo->free_blocks = config.nr_blocks;
714         sbinfo->max_inodes = config.nr_inodes;
715         sbinfo->free_inodes = config.nr_inodes;
716         sb->s_maxbytes = MAX_LFS_FILESIZE;
717         sb->s_blocksize = HPAGE_SIZE;
718         sb->s_blocksize_bits = HPAGE_SHIFT;
719         sb->s_magic = HUGETLBFS_MAGIC;
720         sb->s_op = &hugetlbfs_ops;
721         sb->s_time_gran = 1;
722         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
723                                         S_IFDIR | config.mode, 0);
724         if (!inode)
725                 goto out_free;
726
727         root = d_alloc_root(inode);
728         if (!root) {
729                 iput(inode);
730                 goto out_free;
731         }
732         sb->s_root = root;
733         return 0;
734 out_free:
735         kfree(sbinfo);
736         return -ENOMEM;
737 }
738
739 int hugetlb_get_quota(struct address_space *mapping)
740 {
741         int ret = 0;
742         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
743
744         if (sbinfo->free_blocks > -1) {
745                 spin_lock(&sbinfo->stat_lock);
746                 if (sbinfo->free_blocks > 0)
747                         sbinfo->free_blocks--;
748                 else
749                         ret = -ENOMEM;
750                 spin_unlock(&sbinfo->stat_lock);
751         }
752
753         return ret;
754 }
755
756 void hugetlb_put_quota(struct address_space *mapping)
757 {
758         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
759
760         if (sbinfo->free_blocks > -1) {
761                 spin_lock(&sbinfo->stat_lock);
762                 sbinfo->free_blocks++;
763                 spin_unlock(&sbinfo->stat_lock);
764         }
765 }
766
767 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
768         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
769 {
770         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
771 }
772
773 static struct file_system_type hugetlbfs_fs_type = {
774         .name           = "hugetlbfs",
775         .get_sb         = hugetlbfs_get_sb,
776         .kill_sb        = kill_litter_super,
777 };
778
779 static struct vfsmount *hugetlbfs_vfsmount;
780
781 static int can_do_hugetlb_shm(void)
782 {
783         return likely(capable(CAP_IPC_LOCK) ||
784                         in_group_p(sysctl_hugetlb_shm_group) ||
785                         can_do_mlock());
786 }
787
788 struct file *hugetlb_file_setup(const char *name, size_t size)
789 {
790         int error = -ENOMEM;
791         struct file *file;
792         struct inode *inode;
793         struct dentry *dentry, *root;
794         struct qstr quick_string;
795
796         if (!hugetlbfs_vfsmount)
797                 return ERR_PTR(-ENOENT);
798
799         if (!can_do_hugetlb_shm())
800                 return ERR_PTR(-EPERM);
801
802         if (!user_shm_lock(size, current->user))
803                 return ERR_PTR(-ENOMEM);
804
805         root = hugetlbfs_vfsmount->mnt_root;
806         quick_string.name = name;
807         quick_string.len = strlen(quick_string.name);
808         quick_string.hash = 0;
809         dentry = d_alloc(root, &quick_string);
810         if (!dentry)
811                 goto out_shm_unlock;
812
813         error = -ENFILE;
814         file = get_empty_filp();
815         if (!file)
816                 goto out_dentry;
817
818         error = -ENOSPC;
819         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
820                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
821         if (!inode)
822                 goto out_file;
823
824         error = -ENOMEM;
825         if (hugetlb_reserve_pages(inode, 0, size >> HPAGE_SHIFT))
826                 goto out_inode;
827
828         d_instantiate(dentry, inode);
829         inode->i_size = size;
830         inode->i_nlink = 0;
831         file->f_path.mnt = mntget(hugetlbfs_vfsmount);
832         file->f_path.dentry = dentry;
833         file->f_mapping = inode->i_mapping;
834         file->f_op = &hugetlbfs_file_operations;
835         file->f_mode = FMODE_WRITE | FMODE_READ;
836         return file;
837
838 out_inode:
839         iput(inode);
840 out_file:
841         put_filp(file);
842 out_dentry:
843         dput(dentry);
844 out_shm_unlock:
845         user_shm_unlock(size, current->user);
846         return ERR_PTR(error);
847 }
848
849 static int __init init_hugetlbfs_fs(void)
850 {
851         int error;
852         struct vfsmount *vfsmount;
853
854         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
855                                         sizeof(struct hugetlbfs_inode_info),
856                                         0, 0, init_once);
857         if (hugetlbfs_inode_cachep == NULL)
858                 return -ENOMEM;
859
860         error = register_filesystem(&hugetlbfs_fs_type);
861         if (error)
862                 goto out;
863
864         vfsmount = kern_mount(&hugetlbfs_fs_type);
865
866         if (!IS_ERR(vfsmount)) {
867                 hugetlbfs_vfsmount = vfsmount;
868                 return 0;
869         }
870
871         error = PTR_ERR(vfsmount);
872
873  out:
874         if (error)
875                 kmem_cache_destroy(hugetlbfs_inode_cachep);
876         return error;
877 }
878
879 static void __exit exit_hugetlbfs_fs(void)
880 {
881         kmem_cache_destroy(hugetlbfs_inode_cachep);
882         unregister_filesystem(&hugetlbfs_fs_type);
883 }
884
885 module_init(init_hugetlbfs_fs)
886 module_exit(exit_hugetlbfs_fs)
887
888 MODULE_LICENSE("GPL");