[PATCH] VFS: Permit filesystem to override root dentry on mount
[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/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/capability.h>
22 #include <linux/backing-dev.h>
23 #include <linux/hugetlb.h>
24 #include <linux/pagevec.h>
25 #include <linux/quotaops.h>
26 #include <linux/slab.h>
27 #include <linux/dnotify.h>
28 #include <linux/statfs.h>
29 #include <linux/security.h>
30
31 #include <asm/uaccess.h>
32
33 /* some random number */
34 #define HUGETLBFS_MAGIC 0x958458f6
35
36 static struct super_operations hugetlbfs_ops;
37 static struct address_space_operations hugetlbfs_aops;
38 const struct file_operations hugetlbfs_file_operations;
39 static struct inode_operations hugetlbfs_dir_inode_operations;
40 static struct inode_operations hugetlbfs_inode_operations;
41
42 static struct backing_dev_info hugetlbfs_backing_dev_info = {
43         .ra_pages       = 0,    /* No readahead */
44         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
45 };
46
47 int sysctl_hugetlb_shm_group;
48
49 static void huge_pagevec_release(struct pagevec *pvec)
50 {
51         int i;
52
53         for (i = 0; i < pagevec_count(pvec); ++i)
54                 put_page(pvec->pages[i]);
55
56         pagevec_reinit(pvec);
57 }
58
59 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
60 {
61         struct inode *inode = file->f_dentry->d_inode;
62         struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
63         loff_t len, vma_len;
64         int ret;
65
66         if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
67                 return -EINVAL;
68
69         if (vma->vm_start & ~HPAGE_MASK)
70                 return -EINVAL;
71
72         if (vma->vm_end & ~HPAGE_MASK)
73                 return -EINVAL;
74
75         if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
76                 return -EINVAL;
77
78         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
79
80         mutex_lock(&inode->i_mutex);
81         file_accessed(file);
82         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
83         vma->vm_ops = &hugetlb_vm_ops;
84
85         ret = -ENOMEM;
86         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
87         if (!(vma->vm_flags & VM_WRITE) && len > inode->i_size)
88                 goto out;
89
90         if (vma->vm_flags & VM_MAYSHARE)
91                 if (hugetlb_extend_reservation(info, len >> HPAGE_SHIFT) != 0)
92                         goto out;
93
94         ret = 0;
95         hugetlb_prefault_arch_hook(vma->vm_mm);
96         if (inode->i_size < len)
97                 inode->i_size = len;
98 out:
99         mutex_unlock(&inode->i_mutex);
100
101         return ret;
102 }
103
104 /*
105  * Called under down_write(mmap_sem).
106  */
107
108 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
109 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
110                 unsigned long len, unsigned long pgoff, unsigned long flags);
111 #else
112 static unsigned long
113 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
114                 unsigned long len, unsigned long pgoff, unsigned long flags)
115 {
116         struct mm_struct *mm = current->mm;
117         struct vm_area_struct *vma;
118         unsigned long start_addr;
119
120         if (len & ~HPAGE_MASK)
121                 return -EINVAL;
122         if (len > TASK_SIZE)
123                 return -ENOMEM;
124
125         if (addr) {
126                 addr = ALIGN(addr, HPAGE_SIZE);
127                 vma = find_vma(mm, addr);
128                 if (TASK_SIZE - len >= addr &&
129                     (!vma || addr + len <= vma->vm_start))
130                         return addr;
131         }
132
133         start_addr = mm->free_area_cache;
134
135         if (len <= mm->cached_hole_size)
136                 start_addr = TASK_UNMAPPED_BASE;
137
138 full_search:
139         addr = ALIGN(start_addr, HPAGE_SIZE);
140
141         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
142                 /* At this point:  (!vma || addr < vma->vm_end). */
143                 if (TASK_SIZE - len < addr) {
144                         /*
145                          * Start a new search - just in case we missed
146                          * some holes.
147                          */
148                         if (start_addr != TASK_UNMAPPED_BASE) {
149                                 start_addr = TASK_UNMAPPED_BASE;
150                                 goto full_search;
151                         }
152                         return -ENOMEM;
153                 }
154
155                 if (!vma || addr + len <= vma->vm_start)
156                         return addr;
157                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
158         }
159 }
160 #endif
161
162 /*
163  * Read a page. Again trivial. If it didn't already exist
164  * in the page cache, it is zero-filled.
165  */
166 static int hugetlbfs_readpage(struct file *file, struct page * page)
167 {
168         unlock_page(page);
169         return -EINVAL;
170 }
171
172 static int hugetlbfs_prepare_write(struct file *file,
173                         struct page *page, unsigned offset, unsigned to)
174 {
175         return -EINVAL;
176 }
177
178 static int hugetlbfs_commit_write(struct file *file,
179                         struct page *page, unsigned offset, unsigned to)
180 {
181         return -EINVAL;
182 }
183
184 static void truncate_huge_page(struct page *page)
185 {
186         clear_page_dirty(page);
187         ClearPageUptodate(page);
188         remove_from_page_cache(page);
189         put_page(page);
190 }
191
192 static void truncate_hugepages(struct inode *inode, loff_t lstart)
193 {
194         struct address_space *mapping = &inode->i_data;
195         const pgoff_t start = lstart >> HPAGE_SHIFT;
196         struct pagevec pvec;
197         pgoff_t next;
198         int i;
199
200         hugetlb_truncate_reservation(HUGETLBFS_I(inode),
201                                      lstart >> HPAGE_SHIFT);
202         if (!mapping->nrpages)
203                 return;
204         pagevec_init(&pvec, 0);
205         next = start;
206         while (1) {
207                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
208                         if (next == start)
209                                 break;
210                         next = start;
211                         continue;
212                 }
213
214                 for (i = 0; i < pagevec_count(&pvec); ++i) {
215                         struct page *page = pvec.pages[i];
216
217                         lock_page(page);
218                         if (page->index > next)
219                                 next = page->index;
220                         ++next;
221                         truncate_huge_page(page);
222                         unlock_page(page);
223                         hugetlb_put_quota(mapping);
224                 }
225                 huge_pagevec_release(&pvec);
226         }
227         BUG_ON(!lstart && mapping->nrpages);
228 }
229
230 static void hugetlbfs_delete_inode(struct inode *inode)
231 {
232         truncate_hugepages(inode, 0);
233         clear_inode(inode);
234 }
235
236 static void hugetlbfs_forget_inode(struct inode *inode)
237 {
238         struct super_block *sb = inode->i_sb;
239
240         if (!hlist_unhashed(&inode->i_hash)) {
241                 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
242                         list_move(&inode->i_list, &inode_unused);
243                 inodes_stat.nr_unused++;
244                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
245                         spin_unlock(&inode_lock);
246                         return;
247                 }
248                 inode->i_state |= I_WILL_FREE;
249                 spin_unlock(&inode_lock);
250                 /*
251                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
252                  * in our backing_dev_info.
253                  */
254                 write_inode_now(inode, 1);
255                 spin_lock(&inode_lock);
256                 inode->i_state &= ~I_WILL_FREE;
257                 inodes_stat.nr_unused--;
258                 hlist_del_init(&inode->i_hash);
259         }
260         list_del_init(&inode->i_list);
261         list_del_init(&inode->i_sb_list);
262         inode->i_state |= I_FREEING;
263         inodes_stat.nr_inodes--;
264         spin_unlock(&inode_lock);
265         truncate_hugepages(inode, 0);
266         clear_inode(inode);
267         destroy_inode(inode);
268 }
269
270 static void hugetlbfs_drop_inode(struct inode *inode)
271 {
272         if (!inode->i_nlink)
273                 generic_delete_inode(inode);
274         else
275                 hugetlbfs_forget_inode(inode);
276 }
277
278 /*
279  * h_pgoff is in HPAGE_SIZE units.
280  * vma->vm_pgoff is in PAGE_SIZE units.
281  */
282 static inline void
283 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
284 {
285         struct vm_area_struct *vma;
286         struct prio_tree_iter iter;
287
288         vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
289                 unsigned long h_vm_pgoff;
290                 unsigned long v_offset;
291
292                 h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
293                 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
294                 /*
295                  * Is this VMA fully outside the truncation point?
296                  */
297                 if (h_vm_pgoff >= h_pgoff)
298                         v_offset = 0;
299
300                 unmap_hugepage_range(vma,
301                                 vma->vm_start + v_offset, vma->vm_end);
302         }
303 }
304
305 /*
306  * Expanding truncates are not allowed.
307  */
308 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
309 {
310         unsigned long pgoff;
311         struct address_space *mapping = inode->i_mapping;
312
313         if (offset > inode->i_size)
314                 return -EINVAL;
315
316         BUG_ON(offset & ~HPAGE_MASK);
317         pgoff = offset >> HPAGE_SHIFT;
318
319         inode->i_size = offset;
320         spin_lock(&mapping->i_mmap_lock);
321         if (!prio_tree_empty(&mapping->i_mmap))
322                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
323         spin_unlock(&mapping->i_mmap_lock);
324         truncate_hugepages(inode, offset);
325         return 0;
326 }
327
328 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
329 {
330         struct inode *inode = dentry->d_inode;
331         int error;
332         unsigned int ia_valid = attr->ia_valid;
333
334         BUG_ON(!inode);
335
336         error = inode_change_ok(inode, attr);
337         if (error)
338                 goto out;
339
340         if (ia_valid & ATTR_SIZE) {
341                 error = -EINVAL;
342                 if (!(attr->ia_size & ~HPAGE_MASK))
343                         error = hugetlb_vmtruncate(inode, attr->ia_size);
344                 if (error)
345                         goto out;
346                 attr->ia_valid &= ~ATTR_SIZE;
347         }
348         error = inode_setattr(inode, attr);
349 out:
350         return error;
351 }
352
353 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid, 
354                                         gid_t gid, int mode, dev_t dev)
355 {
356         struct inode *inode;
357
358         inode = new_inode(sb);
359         if (inode) {
360                 struct hugetlbfs_inode_info *info;
361                 inode->i_mode = mode;
362                 inode->i_uid = uid;
363                 inode->i_gid = gid;
364                 inode->i_blksize = HPAGE_SIZE;
365                 inode->i_blocks = 0;
366                 inode->i_mapping->a_ops = &hugetlbfs_aops;
367                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
368                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
369                 info = HUGETLBFS_I(inode);
370                 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
371                 switch (mode & S_IFMT) {
372                 default:
373                         init_special_inode(inode, mode, dev);
374                         break;
375                 case S_IFREG:
376                         inode->i_op = &hugetlbfs_inode_operations;
377                         inode->i_fop = &hugetlbfs_file_operations;
378                         break;
379                 case S_IFDIR:
380                         inode->i_op = &hugetlbfs_dir_inode_operations;
381                         inode->i_fop = &simple_dir_operations;
382
383                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
384                         inode->i_nlink++;
385                         break;
386                 case S_IFLNK:
387                         inode->i_op = &page_symlink_inode_operations;
388                         break;
389                 }
390         }
391         return inode;
392 }
393
394 /*
395  * File creation. Allocate an inode, and we're done..
396  */
397 static int hugetlbfs_mknod(struct inode *dir,
398                         struct dentry *dentry, int mode, dev_t dev)
399 {
400         struct inode *inode;
401         int error = -ENOSPC;
402         gid_t gid;
403
404         if (dir->i_mode & S_ISGID) {
405                 gid = dir->i_gid;
406                 if (S_ISDIR(mode))
407                         mode |= S_ISGID;
408         } else {
409                 gid = current->fsgid;
410         }
411         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
412         if (inode) {
413                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
414                 d_instantiate(dentry, inode);
415                 dget(dentry);   /* Extra count - pin the dentry in core */
416                 error = 0;
417         }
418         return error;
419 }
420
421 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
422 {
423         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
424         if (!retval)
425                 dir->i_nlink++;
426         return retval;
427 }
428
429 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
430 {
431         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
432 }
433
434 static int hugetlbfs_symlink(struct inode *dir,
435                         struct dentry *dentry, const char *symname)
436 {
437         struct inode *inode;
438         int error = -ENOSPC;
439         gid_t gid;
440
441         if (dir->i_mode & S_ISGID)
442                 gid = dir->i_gid;
443         else
444                 gid = current->fsgid;
445
446         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
447                                         gid, S_IFLNK|S_IRWXUGO, 0);
448         if (inode) {
449                 int l = strlen(symname)+1;
450                 error = page_symlink(inode, symname, l);
451                 if (!error) {
452                         d_instantiate(dentry, inode);
453                         dget(dentry);
454                 } else
455                         iput(inode);
456         }
457         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
458
459         return error;
460 }
461
462 /*
463  * For direct-IO reads into hugetlb pages
464  */
465 static int hugetlbfs_set_page_dirty(struct page *page)
466 {
467         return 0;
468 }
469
470 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
471 {
472         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
473
474         buf->f_type = HUGETLBFS_MAGIC;
475         buf->f_bsize = HPAGE_SIZE;
476         if (sbinfo) {
477                 spin_lock(&sbinfo->stat_lock);
478                 /* If no limits set, just report 0 for max/free/used
479                  * blocks, like simple_statfs() */
480                 if (sbinfo->max_blocks >= 0) {
481                         buf->f_blocks = sbinfo->max_blocks;
482                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
483                         buf->f_files = sbinfo->max_inodes;
484                         buf->f_ffree = sbinfo->free_inodes;
485                 }
486                 spin_unlock(&sbinfo->stat_lock);
487         }
488         buf->f_namelen = NAME_MAX;
489         return 0;
490 }
491
492 static void hugetlbfs_put_super(struct super_block *sb)
493 {
494         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
495
496         if (sbi) {
497                 sb->s_fs_info = NULL;
498                 kfree(sbi);
499         }
500 }
501
502 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
503 {
504         if (sbinfo->free_inodes >= 0) {
505                 spin_lock(&sbinfo->stat_lock);
506                 if (unlikely(!sbinfo->free_inodes)) {
507                         spin_unlock(&sbinfo->stat_lock);
508                         return 0;
509                 }
510                 sbinfo->free_inodes--;
511                 spin_unlock(&sbinfo->stat_lock);
512         }
513
514         return 1;
515 }
516
517 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
518 {
519         if (sbinfo->free_inodes >= 0) {
520                 spin_lock(&sbinfo->stat_lock);
521                 sbinfo->free_inodes++;
522                 spin_unlock(&sbinfo->stat_lock);
523         }
524 }
525
526
527 static kmem_cache_t *hugetlbfs_inode_cachep;
528
529 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
530 {
531         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
532         struct hugetlbfs_inode_info *p;
533
534         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
535                 return NULL;
536         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
537         if (unlikely(!p)) {
538                 hugetlbfs_inc_free_inodes(sbinfo);
539                 return NULL;
540         }
541         p->prereserved_hpages = 0;
542         return &p->vfs_inode;
543 }
544
545 static void hugetlbfs_destroy_inode(struct inode *inode)
546 {
547         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
548         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
549         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
550 }
551
552 static struct address_space_operations hugetlbfs_aops = {
553         .readpage       = hugetlbfs_readpage,
554         .prepare_write  = hugetlbfs_prepare_write,
555         .commit_write   = hugetlbfs_commit_write,
556         .set_page_dirty = hugetlbfs_set_page_dirty,
557 };
558
559
560 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
561 {
562         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
563
564         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
565             SLAB_CTOR_CONSTRUCTOR)
566                 inode_init_once(&ei->vfs_inode);
567 }
568
569 const struct file_operations hugetlbfs_file_operations = {
570         .mmap                   = hugetlbfs_file_mmap,
571         .fsync                  = simple_sync_file,
572         .get_unmapped_area      = hugetlb_get_unmapped_area,
573 };
574
575 static struct inode_operations hugetlbfs_dir_inode_operations = {
576         .create         = hugetlbfs_create,
577         .lookup         = simple_lookup,
578         .link           = simple_link,
579         .unlink         = simple_unlink,
580         .symlink        = hugetlbfs_symlink,
581         .mkdir          = hugetlbfs_mkdir,
582         .rmdir          = simple_rmdir,
583         .mknod          = hugetlbfs_mknod,
584         .rename         = simple_rename,
585         .setattr        = hugetlbfs_setattr,
586 };
587
588 static struct inode_operations hugetlbfs_inode_operations = {
589         .setattr        = hugetlbfs_setattr,
590 };
591
592 static struct super_operations hugetlbfs_ops = {
593         .alloc_inode    = hugetlbfs_alloc_inode,
594         .destroy_inode  = hugetlbfs_destroy_inode,
595         .statfs         = hugetlbfs_statfs,
596         .delete_inode   = hugetlbfs_delete_inode,
597         .drop_inode     = hugetlbfs_drop_inode,
598         .put_super      = hugetlbfs_put_super,
599 };
600
601 static int
602 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
603 {
604         char *opt, *value, *rest;
605
606         if (!options)
607                 return 0;
608         while ((opt = strsep(&options, ",")) != NULL) {
609                 if (!*opt)
610                         continue;
611
612                 value = strchr(opt, '=');
613                 if (!value || !*value)
614                         return -EINVAL;
615                 else
616                         *value++ = '\0';
617
618                 if (!strcmp(opt, "uid"))
619                         pconfig->uid = simple_strtoul(value, &value, 0);
620                 else if (!strcmp(opt, "gid"))
621                         pconfig->gid = simple_strtoul(value, &value, 0);
622                 else if (!strcmp(opt, "mode"))
623                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
624                 else if (!strcmp(opt, "size")) {
625                         unsigned long long size = memparse(value, &rest);
626                         if (*rest == '%') {
627                                 size <<= HPAGE_SHIFT;
628                                 size *= max_huge_pages;
629                                 do_div(size, 100);
630                                 rest++;
631                         }
632                         size &= HPAGE_MASK;
633                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
634                         value = rest;
635                 } else if (!strcmp(opt,"nr_inodes")) {
636                         pconfig->nr_inodes = memparse(value, &rest);
637                         value = rest;
638                 } else
639                         return -EINVAL;
640
641                 if (*value)
642                         return -EINVAL;
643         }
644         return 0;
645 }
646
647 static int
648 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
649 {
650         struct inode * inode;
651         struct dentry * root;
652         int ret;
653         struct hugetlbfs_config config;
654         struct hugetlbfs_sb_info *sbinfo;
655
656         config.nr_blocks = -1; /* No limit on size by default */
657         config.nr_inodes = -1; /* No limit on number of inodes by default */
658         config.uid = current->fsuid;
659         config.gid = current->fsgid;
660         config.mode = 0755;
661         ret = hugetlbfs_parse_options(data, &config);
662
663         if (ret)
664                 return ret;
665
666         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
667         if (!sbinfo)
668                 return -ENOMEM;
669         sb->s_fs_info = sbinfo;
670         spin_lock_init(&sbinfo->stat_lock);
671         sbinfo->max_blocks = config.nr_blocks;
672         sbinfo->free_blocks = config.nr_blocks;
673         sbinfo->max_inodes = config.nr_inodes;
674         sbinfo->free_inodes = config.nr_inodes;
675         sb->s_maxbytes = MAX_LFS_FILESIZE;
676         sb->s_blocksize = HPAGE_SIZE;
677         sb->s_blocksize_bits = HPAGE_SHIFT;
678         sb->s_magic = HUGETLBFS_MAGIC;
679         sb->s_op = &hugetlbfs_ops;
680         sb->s_time_gran = 1;
681         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
682                                         S_IFDIR | config.mode, 0);
683         if (!inode)
684                 goto out_free;
685
686         root = d_alloc_root(inode);
687         if (!root) {
688                 iput(inode);
689                 goto out_free;
690         }
691         sb->s_root = root;
692         return 0;
693 out_free:
694         kfree(sbinfo);
695         return -ENOMEM;
696 }
697
698 int hugetlb_get_quota(struct address_space *mapping)
699 {
700         int ret = 0;
701         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
702
703         if (sbinfo->free_blocks > -1) {
704                 spin_lock(&sbinfo->stat_lock);
705                 if (sbinfo->free_blocks > 0)
706                         sbinfo->free_blocks--;
707                 else
708                         ret = -ENOMEM;
709                 spin_unlock(&sbinfo->stat_lock);
710         }
711
712         return ret;
713 }
714
715 void hugetlb_put_quota(struct address_space *mapping)
716 {
717         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
718
719         if (sbinfo->free_blocks > -1) {
720                 spin_lock(&sbinfo->stat_lock);
721                 sbinfo->free_blocks++;
722                 spin_unlock(&sbinfo->stat_lock);
723         }
724 }
725
726 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
727         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
728 {
729         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
730 }
731
732 static struct file_system_type hugetlbfs_fs_type = {
733         .name           = "hugetlbfs",
734         .get_sb         = hugetlbfs_get_sb,
735         .kill_sb        = kill_litter_super,
736 };
737
738 static struct vfsmount *hugetlbfs_vfsmount;
739
740 static int can_do_hugetlb_shm(void)
741 {
742         return likely(capable(CAP_IPC_LOCK) ||
743                         in_group_p(sysctl_hugetlb_shm_group) ||
744                         can_do_mlock());
745 }
746
747 struct file *hugetlb_zero_setup(size_t size)
748 {
749         int error = -ENOMEM;
750         struct file *file;
751         struct inode *inode;
752         struct dentry *dentry, *root;
753         struct qstr quick_string;
754         char buf[16];
755         static atomic_t counter;
756
757         if (!can_do_hugetlb_shm())
758                 return ERR_PTR(-EPERM);
759
760         if (!user_shm_lock(size, current->user))
761                 return ERR_PTR(-ENOMEM);
762
763         root = hugetlbfs_vfsmount->mnt_root;
764         snprintf(buf, 16, "%u", atomic_inc_return(&counter));
765         quick_string.name = buf;
766         quick_string.len = strlen(quick_string.name);
767         quick_string.hash = 0;
768         dentry = d_alloc(root, &quick_string);
769         if (!dentry)
770                 goto out_shm_unlock;
771
772         error = -ENFILE;
773         file = get_empty_filp();
774         if (!file)
775                 goto out_dentry;
776
777         error = -ENOSPC;
778         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
779                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
780         if (!inode)
781                 goto out_file;
782
783         error = -ENOMEM;
784         if (hugetlb_extend_reservation(HUGETLBFS_I(inode),
785                                        size >> HPAGE_SHIFT) != 0)
786                 goto out_inode;
787
788         d_instantiate(dentry, inode);
789         inode->i_size = size;
790         inode->i_nlink = 0;
791         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
792         file->f_dentry = dentry;
793         file->f_mapping = inode->i_mapping;
794         file->f_op = &hugetlbfs_file_operations;
795         file->f_mode = FMODE_WRITE | FMODE_READ;
796         return file;
797
798 out_inode:
799         iput(inode);
800 out_file:
801         put_filp(file);
802 out_dentry:
803         dput(dentry);
804 out_shm_unlock:
805         user_shm_unlock(size, current->user);
806         return ERR_PTR(error);
807 }
808
809 static int __init init_hugetlbfs_fs(void)
810 {
811         int error;
812         struct vfsmount *vfsmount;
813
814         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
815                                         sizeof(struct hugetlbfs_inode_info),
816                                         0, 0, init_once, NULL);
817         if (hugetlbfs_inode_cachep == NULL)
818                 return -ENOMEM;
819
820         error = register_filesystem(&hugetlbfs_fs_type);
821         if (error)
822                 goto out;
823
824         vfsmount = kern_mount(&hugetlbfs_fs_type);
825
826         if (!IS_ERR(vfsmount)) {
827                 hugetlbfs_vfsmount = vfsmount;
828                 return 0;
829         }
830
831         error = PTR_ERR(vfsmount);
832
833  out:
834         if (error)
835                 kmem_cache_destroy(hugetlbfs_inode_cachep);
836         return error;
837 }
838
839 static void __exit exit_hugetlbfs_fs(void)
840 {
841         kmem_cache_destroy(hugetlbfs_inode_cachep);
842         unregister_filesystem(&hugetlbfs_fs_type);
843 }
844
845 module_init(init_hugetlbfs_fs)
846 module_exit(exit_hugetlbfs_fs)
847
848 MODULE_LICENSE("GPL");