get_unmapped_area handles MAP_FIXED in hugetlbfs
[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/mman.h>
26 #include <linux/quotaops.h>
27 #include <linux/slab.h>
28 #include <linux/dnotify.h>
29 #include <linux/statfs.h>
30 #include <linux/security.h>
31
32 #include <asm/uaccess.h>
33
34 /* some random number */
35 #define HUGETLBFS_MAGIC 0x958458f6
36
37 static const struct super_operations hugetlbfs_ops;
38 static const struct address_space_operations hugetlbfs_aops;
39 const struct file_operations hugetlbfs_file_operations;
40 static const struct inode_operations hugetlbfs_dir_inode_operations;
41 static const struct inode_operations hugetlbfs_inode_operations;
42
43 static struct backing_dev_info hugetlbfs_backing_dev_info = {
44         .ra_pages       = 0,    /* No readahead */
45         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
46 };
47
48 int sysctl_hugetlb_shm_group;
49
50 static void huge_pagevec_release(struct pagevec *pvec)
51 {
52         int i;
53
54         for (i = 0; i < pagevec_count(pvec); ++i)
55                 put_page(pvec->pages[i]);
56
57         pagevec_reinit(pvec);
58 }
59
60 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
61 {
62         struct inode *inode = file->f_path.dentry->d_inode;
63         loff_t len, vma_len;
64         int ret;
65
66         /*
67          * vma alignment has already been checked by prepare_hugepage_range.
68          * If you add any error returns here, do so after setting VM_HUGETLB,
69          * so is_vm_hugetlb_page tests below unmap_region go the right way
70          * when do_mmap_pgoff unwinds (may be important on powerpc and ia64).
71          */
72         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
73         vma->vm_ops = &hugetlb_vm_ops;
74
75         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
76
77         mutex_lock(&inode->i_mutex);
78         file_accessed(file);
79
80         ret = -ENOMEM;
81         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
82
83         if (vma->vm_flags & VM_MAYSHARE &&
84             hugetlb_reserve_pages(inode, vma->vm_pgoff >> (HPAGE_SHIFT-PAGE_SHIFT),
85                                   len >> HPAGE_SHIFT))
86                 goto out;
87
88         ret = 0;
89         hugetlb_prefault_arch_hook(vma->vm_mm);
90         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
91                 inode->i_size = len;
92 out:
93         mutex_unlock(&inode->i_mutex);
94
95         return ret;
96 }
97
98 /*
99  * Called under down_write(mmap_sem).
100  */
101
102 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
103 static unsigned long
104 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
105                 unsigned long len, unsigned long pgoff, unsigned long flags)
106 {
107         struct mm_struct *mm = current->mm;
108         struct vm_area_struct *vma;
109         unsigned long start_addr;
110
111         if (len & ~HPAGE_MASK)
112                 return -EINVAL;
113         if (len > TASK_SIZE)
114                 return -ENOMEM;
115
116         if (flags & MAP_FIXED) {
117                 if (prepare_hugepage_range(addr, len, pgoff))
118                         return -EINVAL;
119                 return addr;
120         }
121
122         if (addr) {
123                 addr = ALIGN(addr, HPAGE_SIZE);
124                 vma = find_vma(mm, addr);
125                 if (TASK_SIZE - len >= addr &&
126                     (!vma || addr + len <= vma->vm_start))
127                         return addr;
128         }
129
130         start_addr = mm->free_area_cache;
131
132         if (len <= mm->cached_hole_size)
133                 start_addr = TASK_UNMAPPED_BASE;
134
135 full_search:
136         addr = ALIGN(start_addr, HPAGE_SIZE);
137
138         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
139                 /* At this point:  (!vma || addr < vma->vm_end). */
140                 if (TASK_SIZE - len < addr) {
141                         /*
142                          * Start a new search - just in case we missed
143                          * some holes.
144                          */
145                         if (start_addr != TASK_UNMAPPED_BASE) {
146                                 start_addr = TASK_UNMAPPED_BASE;
147                                 goto full_search;
148                         }
149                         return -ENOMEM;
150                 }
151
152                 if (!vma || addr + len <= vma->vm_start)
153                         return addr;
154                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
155         }
156 }
157 #endif
158
159 /*
160  * Read a page. Again trivial. If it didn't already exist
161  * in the page cache, it is zero-filled.
162  */
163 static int hugetlbfs_readpage(struct file *file, struct page * page)
164 {
165         unlock_page(page);
166         return -EINVAL;
167 }
168
169 static int hugetlbfs_prepare_write(struct file *file,
170                         struct page *page, unsigned offset, unsigned to)
171 {
172         return -EINVAL;
173 }
174
175 static int hugetlbfs_commit_write(struct file *file,
176                         struct page *page, unsigned offset, unsigned to)
177 {
178         return -EINVAL;
179 }
180
181 static void truncate_huge_page(struct page *page)
182 {
183         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
184         ClearPageUptodate(page);
185         remove_from_page_cache(page);
186         put_page(page);
187 }
188
189 static void truncate_hugepages(struct inode *inode, loff_t lstart)
190 {
191         struct address_space *mapping = &inode->i_data;
192         const pgoff_t start = lstart >> HPAGE_SHIFT;
193         struct pagevec pvec;
194         pgoff_t next;
195         int i, freed = 0;
196
197         pagevec_init(&pvec, 0);
198         next = start;
199         while (1) {
200                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
201                         if (next == start)
202                                 break;
203                         next = start;
204                         continue;
205                 }
206
207                 for (i = 0; i < pagevec_count(&pvec); ++i) {
208                         struct page *page = pvec.pages[i];
209
210                         lock_page(page);
211                         if (page->index > next)
212                                 next = page->index;
213                         ++next;
214                         truncate_huge_page(page);
215                         unlock_page(page);
216                         hugetlb_put_quota(mapping);
217                         freed++;
218                 }
219                 huge_pagevec_release(&pvec);
220         }
221         BUG_ON(!lstart && mapping->nrpages);
222         hugetlb_unreserve_pages(inode, start, freed);
223 }
224
225 static void hugetlbfs_delete_inode(struct inode *inode)
226 {
227         truncate_hugepages(inode, 0);
228         clear_inode(inode);
229 }
230
231 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
232 {
233         struct super_block *sb = inode->i_sb;
234
235         if (!hlist_unhashed(&inode->i_hash)) {
236                 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
237                         list_move(&inode->i_list, &inode_unused);
238                 inodes_stat.nr_unused++;
239                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
240                         spin_unlock(&inode_lock);
241                         return;
242                 }
243                 inode->i_state |= I_WILL_FREE;
244                 spin_unlock(&inode_lock);
245                 /*
246                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
247                  * in our backing_dev_info.
248                  */
249                 write_inode_now(inode, 1);
250                 spin_lock(&inode_lock);
251                 inode->i_state &= ~I_WILL_FREE;
252                 inodes_stat.nr_unused--;
253                 hlist_del_init(&inode->i_hash);
254         }
255         list_del_init(&inode->i_list);
256         list_del_init(&inode->i_sb_list);
257         inode->i_state |= I_FREEING;
258         inodes_stat.nr_inodes--;
259         spin_unlock(&inode_lock);
260         truncate_hugepages(inode, 0);
261         clear_inode(inode);
262         destroy_inode(inode);
263 }
264
265 static void hugetlbfs_drop_inode(struct inode *inode)
266 {
267         if (!inode->i_nlink)
268                 generic_delete_inode(inode);
269         else
270                 hugetlbfs_forget_inode(inode);
271 }
272
273 static inline void
274 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
275 {
276         struct vm_area_struct *vma;
277         struct prio_tree_iter iter;
278
279         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
280                 unsigned long v_offset;
281
282                 /*
283                  * Can the expression below overflow on 32-bit arches?
284                  * No, because the prio_tree returns us only those vmas
285                  * which overlap the truncated area starting at pgoff,
286                  * and no vma on a 32-bit arch can span beyond the 4GB.
287                  */
288                 if (vma->vm_pgoff < pgoff)
289                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
290                 else
291                         v_offset = 0;
292
293                 __unmap_hugepage_range(vma,
294                                 vma->vm_start + v_offset, vma->vm_end);
295         }
296 }
297
298 /*
299  * Expanding truncates are not allowed.
300  */
301 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
302 {
303         pgoff_t pgoff;
304         struct address_space *mapping = inode->i_mapping;
305
306         if (offset > inode->i_size)
307                 return -EINVAL;
308
309         BUG_ON(offset & ~HPAGE_MASK);
310         pgoff = offset >> PAGE_SHIFT;
311
312         inode->i_size = offset;
313         spin_lock(&mapping->i_mmap_lock);
314         if (!prio_tree_empty(&mapping->i_mmap))
315                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
316         spin_unlock(&mapping->i_mmap_lock);
317         truncate_hugepages(inode, offset);
318         return 0;
319 }
320
321 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
322 {
323         struct inode *inode = dentry->d_inode;
324         int error;
325         unsigned int ia_valid = attr->ia_valid;
326
327         BUG_ON(!inode);
328
329         error = inode_change_ok(inode, attr);
330         if (error)
331                 goto out;
332
333         if (ia_valid & ATTR_SIZE) {
334                 error = -EINVAL;
335                 if (!(attr->ia_size & ~HPAGE_MASK))
336                         error = hugetlb_vmtruncate(inode, attr->ia_size);
337                 if (error)
338                         goto out;
339                 attr->ia_valid &= ~ATTR_SIZE;
340         }
341         error = inode_setattr(inode, attr);
342 out:
343         return error;
344 }
345
346 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid, 
347                                         gid_t gid, int mode, dev_t dev)
348 {
349         struct inode *inode;
350
351         inode = new_inode(sb);
352         if (inode) {
353                 struct hugetlbfs_inode_info *info;
354                 inode->i_mode = mode;
355                 inode->i_uid = uid;
356                 inode->i_gid = gid;
357                 inode->i_blocks = 0;
358                 inode->i_mapping->a_ops = &hugetlbfs_aops;
359                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
360                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
361                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
362                 info = HUGETLBFS_I(inode);
363                 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
364                 switch (mode & S_IFMT) {
365                 default:
366                         init_special_inode(inode, mode, dev);
367                         break;
368                 case S_IFREG:
369                         inode->i_op = &hugetlbfs_inode_operations;
370                         inode->i_fop = &hugetlbfs_file_operations;
371                         break;
372                 case S_IFDIR:
373                         inode->i_op = &hugetlbfs_dir_inode_operations;
374                         inode->i_fop = &simple_dir_operations;
375
376                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
377                         inc_nlink(inode);
378                         break;
379                 case S_IFLNK:
380                         inode->i_op = &page_symlink_inode_operations;
381                         break;
382                 }
383         }
384         return inode;
385 }
386
387 /*
388  * File creation. Allocate an inode, and we're done..
389  */
390 static int hugetlbfs_mknod(struct inode *dir,
391                         struct dentry *dentry, int mode, dev_t dev)
392 {
393         struct inode *inode;
394         int error = -ENOSPC;
395         gid_t gid;
396
397         if (dir->i_mode & S_ISGID) {
398                 gid = dir->i_gid;
399                 if (S_ISDIR(mode))
400                         mode |= S_ISGID;
401         } else {
402                 gid = current->fsgid;
403         }
404         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
405         if (inode) {
406                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
407                 d_instantiate(dentry, inode);
408                 dget(dentry);   /* Extra count - pin the dentry in core */
409                 error = 0;
410         }
411         return error;
412 }
413
414 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
415 {
416         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
417         if (!retval)
418                 inc_nlink(dir);
419         return retval;
420 }
421
422 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
423 {
424         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
425 }
426
427 static int hugetlbfs_symlink(struct inode *dir,
428                         struct dentry *dentry, const char *symname)
429 {
430         struct inode *inode;
431         int error = -ENOSPC;
432         gid_t gid;
433
434         if (dir->i_mode & S_ISGID)
435                 gid = dir->i_gid;
436         else
437                 gid = current->fsgid;
438
439         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
440                                         gid, S_IFLNK|S_IRWXUGO, 0);
441         if (inode) {
442                 int l = strlen(symname)+1;
443                 error = page_symlink(inode, symname, l);
444                 if (!error) {
445                         d_instantiate(dentry, inode);
446                         dget(dentry);
447                 } else
448                         iput(inode);
449         }
450         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
451
452         return error;
453 }
454
455 /*
456  * mark the head page dirty
457  */
458 static int hugetlbfs_set_page_dirty(struct page *page)
459 {
460         struct page *head = compound_head(page);
461
462         SetPageDirty(head);
463         return 0;
464 }
465
466 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
467 {
468         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
469
470         buf->f_type = HUGETLBFS_MAGIC;
471         buf->f_bsize = HPAGE_SIZE;
472         if (sbinfo) {
473                 spin_lock(&sbinfo->stat_lock);
474                 /* If no limits set, just report 0 for max/free/used
475                  * blocks, like simple_statfs() */
476                 if (sbinfo->max_blocks >= 0) {
477                         buf->f_blocks = sbinfo->max_blocks;
478                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
479                         buf->f_files = sbinfo->max_inodes;
480                         buf->f_ffree = sbinfo->free_inodes;
481                 }
482                 spin_unlock(&sbinfo->stat_lock);
483         }
484         buf->f_namelen = NAME_MAX;
485         return 0;
486 }
487
488 static void hugetlbfs_put_super(struct super_block *sb)
489 {
490         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
491
492         if (sbi) {
493                 sb->s_fs_info = NULL;
494                 kfree(sbi);
495         }
496 }
497
498 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
499 {
500         if (sbinfo->free_inodes >= 0) {
501                 spin_lock(&sbinfo->stat_lock);
502                 if (unlikely(!sbinfo->free_inodes)) {
503                         spin_unlock(&sbinfo->stat_lock);
504                         return 0;
505                 }
506                 sbinfo->free_inodes--;
507                 spin_unlock(&sbinfo->stat_lock);
508         }
509
510         return 1;
511 }
512
513 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
514 {
515         if (sbinfo->free_inodes >= 0) {
516                 spin_lock(&sbinfo->stat_lock);
517                 sbinfo->free_inodes++;
518                 spin_unlock(&sbinfo->stat_lock);
519         }
520 }
521
522
523 static struct kmem_cache *hugetlbfs_inode_cachep;
524
525 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
526 {
527         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
528         struct hugetlbfs_inode_info *p;
529
530         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
531                 return NULL;
532         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
533         if (unlikely(!p)) {
534                 hugetlbfs_inc_free_inodes(sbinfo);
535                 return NULL;
536         }
537         return &p->vfs_inode;
538 }
539
540 static void hugetlbfs_destroy_inode(struct inode *inode)
541 {
542         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
543         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
544         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
545 }
546
547 static const struct address_space_operations hugetlbfs_aops = {
548         .readpage       = hugetlbfs_readpage,
549         .prepare_write  = hugetlbfs_prepare_write,
550         .commit_write   = hugetlbfs_commit_write,
551         .set_page_dirty = hugetlbfs_set_page_dirty,
552 };
553
554
555 static void init_once(void *foo, struct kmem_cache *cachep, unsigned long flags)
556 {
557         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
558
559         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
560             SLAB_CTOR_CONSTRUCTOR)
561                 inode_init_once(&ei->vfs_inode);
562 }
563
564 const struct file_operations hugetlbfs_file_operations = {
565         .mmap                   = hugetlbfs_file_mmap,
566         .fsync                  = simple_sync_file,
567         .get_unmapped_area      = hugetlb_get_unmapped_area,
568 };
569
570 static const struct inode_operations hugetlbfs_dir_inode_operations = {
571         .create         = hugetlbfs_create,
572         .lookup         = simple_lookup,
573         .link           = simple_link,
574         .unlink         = simple_unlink,
575         .symlink        = hugetlbfs_symlink,
576         .mkdir          = hugetlbfs_mkdir,
577         .rmdir          = simple_rmdir,
578         .mknod          = hugetlbfs_mknod,
579         .rename         = simple_rename,
580         .setattr        = hugetlbfs_setattr,
581 };
582
583 static const struct inode_operations hugetlbfs_inode_operations = {
584         .setattr        = hugetlbfs_setattr,
585 };
586
587 static const struct super_operations hugetlbfs_ops = {
588         .alloc_inode    = hugetlbfs_alloc_inode,
589         .destroy_inode  = hugetlbfs_destroy_inode,
590         .statfs         = hugetlbfs_statfs,
591         .delete_inode   = hugetlbfs_delete_inode,
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                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
628                         value = rest;
629                 } else if (!strcmp(opt,"nr_inodes")) {
630                         pconfig->nr_inodes = memparse(value, &rest);
631                         value = rest;
632                 } else
633                         return -EINVAL;
634
635                 if (*value)
636                         return -EINVAL;
637         }
638         return 0;
639 }
640
641 static int
642 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
643 {
644         struct inode * inode;
645         struct dentry * root;
646         int ret;
647         struct hugetlbfs_config config;
648         struct hugetlbfs_sb_info *sbinfo;
649
650         config.nr_blocks = -1; /* No limit on size by default */
651         config.nr_inodes = -1; /* No limit on number of inodes by default */
652         config.uid = current->fsuid;
653         config.gid = current->fsgid;
654         config.mode = 0755;
655         ret = hugetlbfs_parse_options(data, &config);
656
657         if (ret)
658                 return ret;
659
660         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
661         if (!sbinfo)
662                 return -ENOMEM;
663         sb->s_fs_info = sbinfo;
664         spin_lock_init(&sbinfo->stat_lock);
665         sbinfo->max_blocks = config.nr_blocks;
666         sbinfo->free_blocks = config.nr_blocks;
667         sbinfo->max_inodes = config.nr_inodes;
668         sbinfo->free_inodes = config.nr_inodes;
669         sb->s_maxbytes = MAX_LFS_FILESIZE;
670         sb->s_blocksize = HPAGE_SIZE;
671         sb->s_blocksize_bits = HPAGE_SHIFT;
672         sb->s_magic = HUGETLBFS_MAGIC;
673         sb->s_op = &hugetlbfs_ops;
674         sb->s_time_gran = 1;
675         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
676                                         S_IFDIR | config.mode, 0);
677         if (!inode)
678                 goto out_free;
679
680         root = d_alloc_root(inode);
681         if (!root) {
682                 iput(inode);
683                 goto out_free;
684         }
685         sb->s_root = root;
686         return 0;
687 out_free:
688         kfree(sbinfo);
689         return -ENOMEM;
690 }
691
692 int hugetlb_get_quota(struct address_space *mapping)
693 {
694         int ret = 0;
695         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
696
697         if (sbinfo->free_blocks > -1) {
698                 spin_lock(&sbinfo->stat_lock);
699                 if (sbinfo->free_blocks > 0)
700                         sbinfo->free_blocks--;
701                 else
702                         ret = -ENOMEM;
703                 spin_unlock(&sbinfo->stat_lock);
704         }
705
706         return ret;
707 }
708
709 void hugetlb_put_quota(struct address_space *mapping)
710 {
711         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
712
713         if (sbinfo->free_blocks > -1) {
714                 spin_lock(&sbinfo->stat_lock);
715                 sbinfo->free_blocks++;
716                 spin_unlock(&sbinfo->stat_lock);
717         }
718 }
719
720 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
721         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
722 {
723         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
724 }
725
726 static struct file_system_type hugetlbfs_fs_type = {
727         .name           = "hugetlbfs",
728         .get_sb         = hugetlbfs_get_sb,
729         .kill_sb        = kill_litter_super,
730 };
731
732 static struct vfsmount *hugetlbfs_vfsmount;
733
734 static int can_do_hugetlb_shm(void)
735 {
736         return likely(capable(CAP_IPC_LOCK) ||
737                         in_group_p(sysctl_hugetlb_shm_group) ||
738                         can_do_mlock());
739 }
740
741 struct file *hugetlb_zero_setup(size_t size)
742 {
743         int error = -ENOMEM;
744         struct file *file;
745         struct inode *inode;
746         struct dentry *dentry, *root;
747         struct qstr quick_string;
748         char buf[16];
749         static atomic_t counter;
750
751         if (!can_do_hugetlb_shm())
752                 return ERR_PTR(-EPERM);
753
754         if (!user_shm_lock(size, current->user))
755                 return ERR_PTR(-ENOMEM);
756
757         root = hugetlbfs_vfsmount->mnt_root;
758         snprintf(buf, 16, "%u", atomic_inc_return(&counter));
759         quick_string.name = buf;
760         quick_string.len = strlen(quick_string.name);
761         quick_string.hash = 0;
762         dentry = d_alloc(root, &quick_string);
763         if (!dentry)
764                 goto out_shm_unlock;
765
766         error = -ENFILE;
767         file = get_empty_filp();
768         if (!file)
769                 goto out_dentry;
770
771         error = -ENOSPC;
772         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
773                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
774         if (!inode)
775                 goto out_file;
776
777         error = -ENOMEM;
778         if (hugetlb_reserve_pages(inode, 0, size >> HPAGE_SHIFT))
779                 goto out_inode;
780
781         d_instantiate(dentry, inode);
782         inode->i_size = size;
783         inode->i_nlink = 0;
784         file->f_path.mnt = mntget(hugetlbfs_vfsmount);
785         file->f_path.dentry = dentry;
786         file->f_mapping = inode->i_mapping;
787         file->f_op = &hugetlbfs_file_operations;
788         file->f_mode = FMODE_WRITE | FMODE_READ;
789         return file;
790
791 out_inode:
792         iput(inode);
793 out_file:
794         put_filp(file);
795 out_dentry:
796         dput(dentry);
797 out_shm_unlock:
798         user_shm_unlock(size, current->user);
799         return ERR_PTR(error);
800 }
801
802 static int __init init_hugetlbfs_fs(void)
803 {
804         int error;
805         struct vfsmount *vfsmount;
806
807         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
808                                         sizeof(struct hugetlbfs_inode_info),
809                                         0, 0, init_once, NULL);
810         if (hugetlbfs_inode_cachep == NULL)
811                 return -ENOMEM;
812
813         error = register_filesystem(&hugetlbfs_fs_type);
814         if (error)
815                 goto out;
816
817         vfsmount = kern_mount(&hugetlbfs_fs_type);
818
819         if (!IS_ERR(vfsmount)) {
820                 hugetlbfs_vfsmount = vfsmount;
821                 return 0;
822         }
823
824         error = PTR_ERR(vfsmount);
825
826  out:
827         if (error)
828                 kmem_cache_destroy(hugetlbfs_inode_cachep);
829         return error;
830 }
831
832 static void __exit exit_hugetlbfs_fs(void)
833 {
834         kmem_cache_destroy(hugetlbfs_inode_cachep);
835         unregister_filesystem(&hugetlbfs_fs_type);
836 }
837
838 module_init(init_hugetlbfs_fs)
839 module_exit(exit_hugetlbfs_fs)
840
841 MODULE_LICENSE("GPL");