Linux-2.6.12-rc2
[linux-3.10.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/writeback.h>
17 #include <linux/pagemap.h>
18 #include <linux/highmem.h>
19 #include <linux/init.h>
20 #include <linux/string.h>
21 #include <linux/backing-dev.h>
22 #include <linux/hugetlb.h>
23 #include <linux/pagevec.h>
24 #include <linux/quotaops.h>
25 #include <linux/slab.h>
26 #include <linux/dnotify.h>
27 #include <linux/statfs.h>
28 #include <linux/security.h>
29
30 #include <asm/uaccess.h>
31
32 /* some random number */
33 #define HUGETLBFS_MAGIC 0x958458f6
34
35 static struct super_operations hugetlbfs_ops;
36 static struct address_space_operations hugetlbfs_aops;
37 struct file_operations hugetlbfs_file_operations;
38 static struct inode_operations hugetlbfs_dir_inode_operations;
39 static struct inode_operations hugetlbfs_inode_operations;
40
41 static struct backing_dev_info hugetlbfs_backing_dev_info = {
42         .ra_pages       = 0,    /* No readahead */
43         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
44 };
45
46 int sysctl_hugetlb_shm_group;
47
48 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
49 {
50         struct inode *inode = file->f_dentry->d_inode;
51         struct address_space *mapping = inode->i_mapping;
52         loff_t len, vma_len;
53         int ret;
54
55         if ((vma->vm_flags & (VM_MAYSHARE | VM_WRITE)) == VM_WRITE)
56                 return -EINVAL;
57
58         if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
59                 return -EINVAL;
60
61         if (vma->vm_start & ~HPAGE_MASK)
62                 return -EINVAL;
63
64         if (vma->vm_end & ~HPAGE_MASK)
65                 return -EINVAL;
66
67         if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
68                 return -EINVAL;
69
70         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
71
72         down(&inode->i_sem);
73         file_accessed(file);
74         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
75         vma->vm_ops = &hugetlb_vm_ops;
76
77         ret = -ENOMEM;
78         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
79         if (!(vma->vm_flags & VM_WRITE) && len > inode->i_size)
80                 goto out;
81
82         ret = hugetlb_prefault(mapping, vma);
83         if (ret)
84                 goto out;
85
86         if (inode->i_size < len)
87                 inode->i_size = len;
88 out:
89         up(&inode->i_sem);
90
91         return ret;
92 }
93
94 /*
95  * Called under down_write(mmap_sem), page_table_lock is not held
96  */
97
98 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
99 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
100                 unsigned long len, unsigned long pgoff, unsigned long flags);
101 #else
102 static unsigned long
103 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
104                 unsigned long len, unsigned long pgoff, unsigned long flags)
105 {
106         struct mm_struct *mm = current->mm;
107         struct vm_area_struct *vma;
108         unsigned long start_addr;
109
110         if (len & ~HPAGE_MASK)
111                 return -EINVAL;
112         if (len > TASK_SIZE)
113                 return -ENOMEM;
114
115         if (addr) {
116                 addr = ALIGN(addr, HPAGE_SIZE);
117                 vma = find_vma(mm, addr);
118                 if (TASK_SIZE - len >= addr &&
119                     (!vma || addr + len <= vma->vm_start))
120                         return addr;
121         }
122
123         start_addr = mm->free_area_cache;
124
125 full_search:
126         addr = ALIGN(start_addr, HPAGE_SIZE);
127
128         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
129                 /* At this point:  (!vma || addr < vma->vm_end). */
130                 if (TASK_SIZE - len < addr) {
131                         /*
132                          * Start a new search - just in case we missed
133                          * some holes.
134                          */
135                         if (start_addr != TASK_UNMAPPED_BASE) {
136                                 start_addr = TASK_UNMAPPED_BASE;
137                                 goto full_search;
138                         }
139                         return -ENOMEM;
140                 }
141
142                 if (!vma || addr + len <= vma->vm_start)
143                         return addr;
144                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
145         }
146 }
147 #endif
148
149 /*
150  * Read a page. Again trivial. If it didn't already exist
151  * in the page cache, it is zero-filled.
152  */
153 static int hugetlbfs_readpage(struct file *file, struct page * page)
154 {
155         unlock_page(page);
156         return -EINVAL;
157 }
158
159 static int hugetlbfs_prepare_write(struct file *file,
160                         struct page *page, unsigned offset, unsigned to)
161 {
162         return -EINVAL;
163 }
164
165 static int hugetlbfs_commit_write(struct file *file,
166                         struct page *page, unsigned offset, unsigned to)
167 {
168         return -EINVAL;
169 }
170
171 static void huge_pagevec_release(struct pagevec *pvec)
172 {
173         int i;
174
175         for (i = 0; i < pagevec_count(pvec); ++i)
176                 put_page(pvec->pages[i]);
177
178         pagevec_reinit(pvec);
179 }
180
181 static void truncate_huge_page(struct page *page)
182 {
183         clear_page_dirty(page);
184         ClearPageUptodate(page);
185         remove_from_page_cache(page);
186         put_page(page);
187 }
188
189 static void truncate_hugepages(struct address_space *mapping, loff_t lstart)
190 {
191         const pgoff_t start = lstart >> HPAGE_SHIFT;
192         struct pagevec pvec;
193         pgoff_t next;
194         int i;
195
196         pagevec_init(&pvec, 0);
197         next = start;
198         while (1) {
199                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
200                         if (next == start)
201                                 break;
202                         next = start;
203                         continue;
204                 }
205
206                 for (i = 0; i < pagevec_count(&pvec); ++i) {
207                         struct page *page = pvec.pages[i];
208
209                         lock_page(page);
210                         if (page->index > next)
211                                 next = page->index;
212                         ++next;
213                         truncate_huge_page(page);
214                         unlock_page(page);
215                         hugetlb_put_quota(mapping);
216                 }
217                 huge_pagevec_release(&pvec);
218         }
219         BUG_ON(!lstart && mapping->nrpages);
220 }
221
222 static void hugetlbfs_delete_inode(struct inode *inode)
223 {
224         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(inode->i_sb);
225
226         hlist_del_init(&inode->i_hash);
227         list_del_init(&inode->i_list);
228         list_del_init(&inode->i_sb_list);
229         inode->i_state |= I_FREEING;
230         inodes_stat.nr_inodes--;
231         spin_unlock(&inode_lock);
232
233         if (inode->i_data.nrpages)
234                 truncate_hugepages(&inode->i_data, 0);
235
236         security_inode_delete(inode);
237
238         if (sbinfo->free_inodes >= 0) {
239                 spin_lock(&sbinfo->stat_lock);
240                 sbinfo->free_inodes++;
241                 spin_unlock(&sbinfo->stat_lock);
242         }
243
244         clear_inode(inode);
245         destroy_inode(inode);
246 }
247
248 static void hugetlbfs_forget_inode(struct inode *inode)
249 {
250         struct super_block *super_block = inode->i_sb;
251         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(super_block);
252
253         if (hlist_unhashed(&inode->i_hash))
254                 goto out_truncate;
255
256         if (!(inode->i_state & (I_DIRTY|I_LOCK))) {
257                 list_del(&inode->i_list);
258                 list_add(&inode->i_list, &inode_unused);
259         }
260         inodes_stat.nr_unused++;
261         if (!super_block || (super_block->s_flags & MS_ACTIVE)) {
262                 spin_unlock(&inode_lock);
263                 return;
264         }
265
266         /* write_inode_now() ? */
267         inodes_stat.nr_unused--;
268         hlist_del_init(&inode->i_hash);
269 out_truncate:
270         list_del_init(&inode->i_list);
271         list_del_init(&inode->i_sb_list);
272         inode->i_state |= I_FREEING;
273         inodes_stat.nr_inodes--;
274         spin_unlock(&inode_lock);
275         if (inode->i_data.nrpages)
276                 truncate_hugepages(&inode->i_data, 0);
277
278         if (sbinfo->free_inodes >= 0) {
279                 spin_lock(&sbinfo->stat_lock);
280                 sbinfo->free_inodes++;
281                 spin_unlock(&sbinfo->stat_lock);
282         }
283
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                 hugetlbfs_delete_inode(inode);
292         else
293                 hugetlbfs_forget_inode(inode);
294 }
295
296 /*
297  * h_pgoff is in HPAGE_SIZE units.
298  * vma->vm_pgoff is in PAGE_SIZE units.
299  */
300 static inline void
301 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
302 {
303         struct vm_area_struct *vma;
304         struct prio_tree_iter iter;
305
306         vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
307                 unsigned long h_vm_pgoff;
308                 unsigned long v_length;
309                 unsigned long v_offset;
310
311                 h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
312                 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
313                 /*
314                  * Is this VMA fully outside the truncation point?
315                  */
316                 if (h_vm_pgoff >= h_pgoff)
317                         v_offset = 0;
318
319                 v_length = vma->vm_end - vma->vm_start;
320
321                 zap_hugepage_range(vma,
322                                 vma->vm_start + v_offset,
323                                 v_length - v_offset);
324         }
325 }
326
327 /*
328  * Expanding truncates are not allowed.
329  */
330 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
331 {
332         unsigned long pgoff;
333         struct address_space *mapping = inode->i_mapping;
334
335         if (offset > inode->i_size)
336                 return -EINVAL;
337
338         BUG_ON(offset & ~HPAGE_MASK);
339         pgoff = offset >> HPAGE_SHIFT;
340
341         inode->i_size = offset;
342         spin_lock(&mapping->i_mmap_lock);
343         if (!prio_tree_empty(&mapping->i_mmap))
344                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
345         spin_unlock(&mapping->i_mmap_lock);
346         truncate_hugepages(mapping, offset);
347         return 0;
348 }
349
350 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
351 {
352         struct inode *inode = dentry->d_inode;
353         int error;
354         unsigned int ia_valid = attr->ia_valid;
355
356         BUG_ON(!inode);
357
358         error = inode_change_ok(inode, attr);
359         if (error)
360                 goto out;
361
362         if (ia_valid & ATTR_SIZE) {
363                 error = -EINVAL;
364                 if (!(attr->ia_size & ~HPAGE_MASK))
365                         error = hugetlb_vmtruncate(inode, attr->ia_size);
366                 if (error)
367                         goto out;
368                 attr->ia_valid &= ~ATTR_SIZE;
369         }
370         error = inode_setattr(inode, attr);
371 out:
372         return error;
373 }
374
375 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid, 
376                                         gid_t gid, int mode, dev_t dev)
377 {
378         struct inode *inode;
379         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
380
381         if (sbinfo->free_inodes >= 0) {
382                 spin_lock(&sbinfo->stat_lock);
383                 if (!sbinfo->free_inodes) {
384                         spin_unlock(&sbinfo->stat_lock);
385                         return NULL;
386                 }
387                 sbinfo->free_inodes--;
388                 spin_unlock(&sbinfo->stat_lock);
389         }
390
391         inode = new_inode(sb);
392         if (inode) {
393                 struct hugetlbfs_inode_info *info;
394                 inode->i_mode = mode;
395                 inode->i_uid = uid;
396                 inode->i_gid = gid;
397                 inode->i_blksize = HPAGE_SIZE;
398                 inode->i_blocks = 0;
399                 inode->i_mapping->a_ops = &hugetlbfs_aops;
400                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
401                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
402                 info = HUGETLBFS_I(inode);
403                 mpol_shared_policy_init(&info->policy);
404                 switch (mode & S_IFMT) {
405                 default:
406                         init_special_inode(inode, mode, dev);
407                         break;
408                 case S_IFREG:
409                         inode->i_op = &hugetlbfs_inode_operations;
410                         inode->i_fop = &hugetlbfs_file_operations;
411                         break;
412                 case S_IFDIR:
413                         inode->i_op = &hugetlbfs_dir_inode_operations;
414                         inode->i_fop = &simple_dir_operations;
415
416                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
417                         inode->i_nlink++;
418                         break;
419                 case S_IFLNK:
420                         inode->i_op = &page_symlink_inode_operations;
421                         break;
422                 }
423         }
424         return inode;
425 }
426
427 /*
428  * File creation. Allocate an inode, and we're done..
429  */
430 static int hugetlbfs_mknod(struct inode *dir,
431                         struct dentry *dentry, int mode, dev_t dev)
432 {
433         struct inode *inode;
434         int error = -ENOSPC;
435         gid_t gid;
436
437         if (dir->i_mode & S_ISGID) {
438                 gid = dir->i_gid;
439                 if (S_ISDIR(mode))
440                         mode |= S_ISGID;
441         } else {
442                 gid = current->fsgid;
443         }
444         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
445         if (inode) {
446                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
447                 d_instantiate(dentry, inode);
448                 dget(dentry);   /* Extra count - pin the dentry in core */
449                 error = 0;
450         }
451         return error;
452 }
453
454 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
455 {
456         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
457         if (!retval)
458                 dir->i_nlink++;
459         return retval;
460 }
461
462 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
463 {
464         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
465 }
466
467 static int hugetlbfs_symlink(struct inode *dir,
468                         struct dentry *dentry, const char *symname)
469 {
470         struct inode *inode;
471         int error = -ENOSPC;
472         gid_t gid;
473
474         if (dir->i_mode & S_ISGID)
475                 gid = dir->i_gid;
476         else
477                 gid = current->fsgid;
478
479         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
480                                         gid, S_IFLNK|S_IRWXUGO, 0);
481         if (inode) {
482                 int l = strlen(symname)+1;
483                 error = page_symlink(inode, symname, l);
484                 if (!error) {
485                         d_instantiate(dentry, inode);
486                         dget(dentry);
487                 } else
488                         iput(inode);
489         }
490         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
491
492         return error;
493 }
494
495 /*
496  * For direct-IO reads into hugetlb pages
497  */
498 static int hugetlbfs_set_page_dirty(struct page *page)
499 {
500         return 0;
501 }
502
503 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
504 {
505         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
506
507         buf->f_type = HUGETLBFS_MAGIC;
508         buf->f_bsize = HPAGE_SIZE;
509         if (sbinfo) {
510                 spin_lock(&sbinfo->stat_lock);
511                 buf->f_blocks = sbinfo->max_blocks;
512                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
513                 buf->f_files = sbinfo->max_inodes;
514                 buf->f_ffree = sbinfo->free_inodes;
515                 spin_unlock(&sbinfo->stat_lock);
516         }
517         buf->f_namelen = NAME_MAX;
518         return 0;
519 }
520
521 static void hugetlbfs_put_super(struct super_block *sb)
522 {
523         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
524
525         if (sbi) {
526                 sb->s_fs_info = NULL;
527                 kfree(sbi);
528         }
529 }
530
531 static kmem_cache_t *hugetlbfs_inode_cachep;
532
533 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
534 {
535         struct hugetlbfs_inode_info *p;
536
537         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
538         if (!p)
539                 return NULL;
540         return &p->vfs_inode;
541 }
542
543 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
544 {
545         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
546
547         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
548             SLAB_CTOR_CONSTRUCTOR)
549                 inode_init_once(&ei->vfs_inode);
550 }
551
552 static void hugetlbfs_destroy_inode(struct inode *inode)
553 {
554         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
555         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
556 }
557
558 static struct address_space_operations hugetlbfs_aops = {
559         .readpage       = hugetlbfs_readpage,
560         .prepare_write  = hugetlbfs_prepare_write,
561         .commit_write   = hugetlbfs_commit_write,
562         .set_page_dirty = hugetlbfs_set_page_dirty,
563 };
564
565 struct file_operations hugetlbfs_file_operations = {
566         .mmap                   = hugetlbfs_file_mmap,
567         .fsync                  = simple_sync_file,
568         .get_unmapped_area      = hugetlb_get_unmapped_area,
569 };
570
571 static struct inode_operations hugetlbfs_dir_inode_operations = {
572         .create         = hugetlbfs_create,
573         .lookup         = simple_lookup,
574         .link           = simple_link,
575         .unlink         = simple_unlink,
576         .symlink        = hugetlbfs_symlink,
577         .mkdir          = hugetlbfs_mkdir,
578         .rmdir          = simple_rmdir,
579         .mknod          = hugetlbfs_mknod,
580         .rename         = simple_rename,
581         .setattr        = hugetlbfs_setattr,
582 };
583
584 static struct inode_operations hugetlbfs_inode_operations = {
585         .setattr        = hugetlbfs_setattr,
586 };
587
588 static struct super_operations hugetlbfs_ops = {
589         .alloc_inode    = hugetlbfs_alloc_inode,
590         .destroy_inode  = hugetlbfs_destroy_inode,
591         .statfs         = hugetlbfs_statfs,
592         .drop_inode     = hugetlbfs_drop_inode,
593         .put_super      = hugetlbfs_put_super,
594 };
595
596 static int
597 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
598 {
599         char *opt, *value, *rest;
600
601         if (!options)
602                 return 0;
603         while ((opt = strsep(&options, ",")) != NULL) {
604                 if (!*opt)
605                         continue;
606
607                 value = strchr(opt, '=');
608                 if (!value || !*value)
609                         return -EINVAL;
610                 else
611                         *value++ = '\0';
612
613                 if (!strcmp(opt, "uid"))
614                         pconfig->uid = simple_strtoul(value, &value, 0);
615                 else if (!strcmp(opt, "gid"))
616                         pconfig->gid = simple_strtoul(value, &value, 0);
617                 else if (!strcmp(opt, "mode"))
618                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
619                 else if (!strcmp(opt, "size")) {
620                         unsigned long long size = memparse(value, &rest);
621                         if (*rest == '%') {
622                                 size <<= HPAGE_SHIFT;
623                                 size *= max_huge_pages;
624                                 do_div(size, 100);
625                                 rest++;
626                         }
627                         size &= HPAGE_MASK;
628                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
629                         value = rest;
630                 } else if (!strcmp(opt,"nr_inodes")) {
631                         pconfig->nr_inodes = memparse(value, &rest);
632                         value = rest;
633                 } else
634                         return -EINVAL;
635
636                 if (*value)
637                         return -EINVAL;
638         }
639         return 0;
640 }
641
642 static int
643 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
644 {
645         struct inode * inode;
646         struct dentry * root;
647         int ret;
648         struct hugetlbfs_config config;
649         struct hugetlbfs_sb_info *sbinfo;
650
651         config.nr_blocks = -1; /* No limit on size by default */
652         config.nr_inodes = -1; /* No limit on number of inodes by default */
653         config.uid = current->fsuid;
654         config.gid = current->fsgid;
655         config.mode = 0755;
656         ret = hugetlbfs_parse_options(data, &config);
657
658         if (ret)
659                 return ret;
660
661         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
662         if (!sbinfo)
663                 return -ENOMEM;
664         sb->s_fs_info = sbinfo;
665         spin_lock_init(&sbinfo->stat_lock);
666         sbinfo->max_blocks = config.nr_blocks;
667         sbinfo->free_blocks = config.nr_blocks;
668         sbinfo->max_inodes = config.nr_inodes;
669         sbinfo->free_inodes = config.nr_inodes;
670         sb->s_maxbytes = MAX_LFS_FILESIZE;
671         sb->s_blocksize = HPAGE_SIZE;
672         sb->s_blocksize_bits = HPAGE_SHIFT;
673         sb->s_magic = HUGETLBFS_MAGIC;
674         sb->s_op = &hugetlbfs_ops;
675         sb->s_time_gran = 1;
676         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
677                                         S_IFDIR | config.mode, 0);
678         if (!inode)
679                 goto out_free;
680
681         root = d_alloc_root(inode);
682         if (!root) {
683                 iput(inode);
684                 goto out_free;
685         }
686         sb->s_root = root;
687         return 0;
688 out_free:
689         kfree(sbinfo);
690         return -ENOMEM;
691 }
692
693 int hugetlb_get_quota(struct address_space *mapping)
694 {
695         int ret = 0;
696         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
697
698         if (sbinfo->free_blocks > -1) {
699                 spin_lock(&sbinfo->stat_lock);
700                 if (sbinfo->free_blocks > 0)
701                         sbinfo->free_blocks--;
702                 else
703                         ret = -ENOMEM;
704                 spin_unlock(&sbinfo->stat_lock);
705         }
706
707         return ret;
708 }
709
710 void hugetlb_put_quota(struct address_space *mapping)
711 {
712         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
713
714         if (sbinfo->free_blocks > -1) {
715                 spin_lock(&sbinfo->stat_lock);
716                 sbinfo->free_blocks++;
717                 spin_unlock(&sbinfo->stat_lock);
718         }
719 }
720
721 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
722         int flags, const char *dev_name, void *data)
723 {
724         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
725 }
726
727 static struct file_system_type hugetlbfs_fs_type = {
728         .name           = "hugetlbfs",
729         .get_sb         = hugetlbfs_get_sb,
730         .kill_sb        = kill_litter_super,
731 };
732
733 static struct vfsmount *hugetlbfs_vfsmount;
734
735 /*
736  * Return the next identifier for a shm file
737  */
738 static unsigned long hugetlbfs_counter(void)
739 {
740         static DEFINE_SPINLOCK(lock);
741         static unsigned long counter;
742         unsigned long ret;
743
744         spin_lock(&lock);
745         ret = ++counter;
746         spin_unlock(&lock);
747         return ret;
748 }
749
750 static int can_do_hugetlb_shm(void)
751 {
752         return likely(capable(CAP_IPC_LOCK) ||
753                         in_group_p(sysctl_hugetlb_shm_group) ||
754                         can_do_mlock());
755 }
756
757 struct file *hugetlb_zero_setup(size_t size)
758 {
759         int error = -ENOMEM;
760         struct file *file;
761         struct inode *inode;
762         struct dentry *dentry, *root;
763         struct qstr quick_string;
764         char buf[16];
765
766         if (!can_do_hugetlb_shm())
767                 return ERR_PTR(-EPERM);
768
769         if (!is_hugepage_mem_enough(size))
770                 return ERR_PTR(-ENOMEM);
771
772         if (!user_shm_lock(size, current->user))
773                 return ERR_PTR(-ENOMEM);
774
775         root = hugetlbfs_vfsmount->mnt_root;
776         snprintf(buf, 16, "%lu", hugetlbfs_counter());
777         quick_string.name = buf;
778         quick_string.len = strlen(quick_string.name);
779         quick_string.hash = 0;
780         dentry = d_alloc(root, &quick_string);
781         if (!dentry)
782                 goto out_shm_unlock;
783
784         error = -ENFILE;
785         file = get_empty_filp();
786         if (!file)
787                 goto out_dentry;
788
789         error = -ENOSPC;
790         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
791                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
792         if (!inode)
793                 goto out_file;
794
795         d_instantiate(dentry, inode);
796         inode->i_size = size;
797         inode->i_nlink = 0;
798         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
799         file->f_dentry = dentry;
800         file->f_mapping = inode->i_mapping;
801         file->f_op = &hugetlbfs_file_operations;
802         file->f_mode = FMODE_WRITE | FMODE_READ;
803         return file;
804
805 out_file:
806         put_filp(file);
807 out_dentry:
808         dput(dentry);
809 out_shm_unlock:
810         user_shm_unlock(size, current->user);
811         return ERR_PTR(error);
812 }
813
814 static int __init init_hugetlbfs_fs(void)
815 {
816         int error;
817         struct vfsmount *vfsmount;
818
819         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
820                                         sizeof(struct hugetlbfs_inode_info),
821                                         0, 0, init_once, NULL);
822         if (hugetlbfs_inode_cachep == NULL)
823                 return -ENOMEM;
824
825         error = register_filesystem(&hugetlbfs_fs_type);
826         if (error)
827                 goto out;
828
829         vfsmount = kern_mount(&hugetlbfs_fs_type);
830
831         if (!IS_ERR(vfsmount)) {
832                 hugetlbfs_vfsmount = vfsmount;
833                 return 0;
834         }
835
836         error = PTR_ERR(vfsmount);
837
838  out:
839         if (error)
840                 kmem_cache_destroy(hugetlbfs_inode_cachep);
841         return error;
842 }
843
844 static void __exit exit_hugetlbfs_fs(void)
845 {
846         kmem_cache_destroy(hugetlbfs_inode_cachep);
847         unregister_filesystem(&hugetlbfs_fs_type);
848 }
849
850 module_init(init_hugetlbfs_fs)
851 module_exit(exit_hugetlbfs_fs)
852
853 MODULE_LICENSE("GPL");