r/o bind mounts: filesystem helpers for custom 'struct file's
[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 static int
183 hugetlbfs_read_actor(struct page *page, unsigned long offset,
184                         char __user *buf, unsigned long count,
185                         unsigned long size)
186 {
187         char *kaddr;
188         unsigned long left, copied = 0;
189         int i, chunksize;
190
191         if (size > count)
192                 size = count;
193
194         /* Find which 4k chunk and offset with in that chunk */
195         i = offset >> PAGE_CACHE_SHIFT;
196         offset = offset & ~PAGE_CACHE_MASK;
197
198         while (size) {
199                 chunksize = PAGE_CACHE_SIZE;
200                 if (offset)
201                         chunksize -= offset;
202                 if (chunksize > size)
203                         chunksize = size;
204                 kaddr = kmap(&page[i]);
205                 left = __copy_to_user(buf, kaddr + offset, chunksize);
206                 kunmap(&page[i]);
207                 if (left) {
208                         copied += (chunksize - left);
209                         break;
210                 }
211                 offset = 0;
212                 size -= chunksize;
213                 buf += chunksize;
214                 copied += chunksize;
215                 i++;
216         }
217         return copied ? copied : -EFAULT;
218 }
219
220 /*
221  * Support for read() - Find the page attached to f_mapping and copy out the
222  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
223  * since it has PAGE_CACHE_SIZE assumptions.
224  */
225 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
226                               size_t len, loff_t *ppos)
227 {
228         struct address_space *mapping = filp->f_mapping;
229         struct inode *inode = mapping->host;
230         unsigned long index = *ppos >> HPAGE_SHIFT;
231         unsigned long offset = *ppos & ~HPAGE_MASK;
232         unsigned long end_index;
233         loff_t isize;
234         ssize_t retval = 0;
235
236         mutex_lock(&inode->i_mutex);
237
238         /* validate length */
239         if (len == 0)
240                 goto out;
241
242         isize = i_size_read(inode);
243         if (!isize)
244                 goto out;
245
246         end_index = (isize - 1) >> HPAGE_SHIFT;
247         for (;;) {
248                 struct page *page;
249                 int nr, ret;
250
251                 /* nr is the maximum number of bytes to copy from this page */
252                 nr = HPAGE_SIZE;
253                 if (index >= end_index) {
254                         if (index > end_index)
255                                 goto out;
256                         nr = ((isize - 1) & ~HPAGE_MASK) + 1;
257                         if (nr <= offset) {
258                                 goto out;
259                         }
260                 }
261                 nr = nr - offset;
262
263                 /* Find the page */
264                 page = find_get_page(mapping, index);
265                 if (unlikely(page == NULL)) {
266                         /*
267                          * We have a HOLE, zero out the user-buffer for the
268                          * length of the hole or request.
269                          */
270                         ret = len < nr ? len : nr;
271                         if (clear_user(buf, ret))
272                                 ret = -EFAULT;
273                 } else {
274                         /*
275                          * We have the page, copy it to user space buffer.
276                          */
277                         ret = hugetlbfs_read_actor(page, offset, buf, len, nr);
278                 }
279                 if (ret < 0) {
280                         if (retval == 0)
281                                 retval = ret;
282                         if (page)
283                                 page_cache_release(page);
284                         goto out;
285                 }
286
287                 offset += ret;
288                 retval += ret;
289                 len -= ret;
290                 index += offset >> HPAGE_SHIFT;
291                 offset &= ~HPAGE_MASK;
292
293                 if (page)
294                         page_cache_release(page);
295
296                 /* short read or no more work */
297                 if ((ret != nr) || (len == 0))
298                         break;
299         }
300 out:
301         *ppos = ((loff_t)index << HPAGE_SHIFT) + offset;
302         mutex_unlock(&inode->i_mutex);
303         return retval;
304 }
305
306 /*
307  * Read a page. Again trivial. If it didn't already exist
308  * in the page cache, it is zero-filled.
309  */
310 static int hugetlbfs_readpage(struct file *file, struct page * page)
311 {
312         unlock_page(page);
313         return -EINVAL;
314 }
315
316 static int hugetlbfs_write_begin(struct file *file,
317                         struct address_space *mapping,
318                         loff_t pos, unsigned len, unsigned flags,
319                         struct page **pagep, void **fsdata)
320 {
321         return -EINVAL;
322 }
323
324 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
325                         loff_t pos, unsigned len, unsigned copied,
326                         struct page *page, void *fsdata)
327 {
328         BUG();
329         return -EINVAL;
330 }
331
332 static void truncate_huge_page(struct page *page)
333 {
334         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
335         ClearPageUptodate(page);
336         remove_from_page_cache(page);
337         put_page(page);
338 }
339
340 static void truncate_hugepages(struct inode *inode, loff_t lstart)
341 {
342         struct address_space *mapping = &inode->i_data;
343         const pgoff_t start = lstart >> HPAGE_SHIFT;
344         struct pagevec pvec;
345         pgoff_t next;
346         int i, freed = 0;
347
348         pagevec_init(&pvec, 0);
349         next = start;
350         while (1) {
351                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
352                         if (next == start)
353                                 break;
354                         next = start;
355                         continue;
356                 }
357
358                 for (i = 0; i < pagevec_count(&pvec); ++i) {
359                         struct page *page = pvec.pages[i];
360
361                         lock_page(page);
362                         if (page->index > next)
363                                 next = page->index;
364                         ++next;
365                         truncate_huge_page(page);
366                         unlock_page(page);
367                         hugetlb_put_quota(mapping);
368                         freed++;
369                 }
370                 huge_pagevec_release(&pvec);
371         }
372         BUG_ON(!lstart && mapping->nrpages);
373         hugetlb_unreserve_pages(inode, start, freed);
374 }
375
376 static void hugetlbfs_delete_inode(struct inode *inode)
377 {
378         truncate_hugepages(inode, 0);
379         clear_inode(inode);
380 }
381
382 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
383 {
384         struct super_block *sb = inode->i_sb;
385
386         if (!hlist_unhashed(&inode->i_hash)) {
387                 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
388                         list_move(&inode->i_list, &inode_unused);
389                 inodes_stat.nr_unused++;
390                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
391                         spin_unlock(&inode_lock);
392                         return;
393                 }
394                 inode->i_state |= I_WILL_FREE;
395                 spin_unlock(&inode_lock);
396                 /*
397                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
398                  * in our backing_dev_info.
399                  */
400                 write_inode_now(inode, 1);
401                 spin_lock(&inode_lock);
402                 inode->i_state &= ~I_WILL_FREE;
403                 inodes_stat.nr_unused--;
404                 hlist_del_init(&inode->i_hash);
405         }
406         list_del_init(&inode->i_list);
407         list_del_init(&inode->i_sb_list);
408         inode->i_state |= I_FREEING;
409         inodes_stat.nr_inodes--;
410         spin_unlock(&inode_lock);
411         truncate_hugepages(inode, 0);
412         clear_inode(inode);
413         destroy_inode(inode);
414 }
415
416 static void hugetlbfs_drop_inode(struct inode *inode)
417 {
418         if (!inode->i_nlink)
419                 generic_delete_inode(inode);
420         else
421                 hugetlbfs_forget_inode(inode);
422 }
423
424 static inline void
425 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
426 {
427         struct vm_area_struct *vma;
428         struct prio_tree_iter iter;
429
430         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
431                 unsigned long v_offset;
432
433                 /*
434                  * Can the expression below overflow on 32-bit arches?
435                  * No, because the prio_tree returns us only those vmas
436                  * which overlap the truncated area starting at pgoff,
437                  * and no vma on a 32-bit arch can span beyond the 4GB.
438                  */
439                 if (vma->vm_pgoff < pgoff)
440                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
441                 else
442                         v_offset = 0;
443
444                 __unmap_hugepage_range(vma,
445                                 vma->vm_start + v_offset, vma->vm_end);
446         }
447 }
448
449 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
450 {
451         pgoff_t pgoff;
452         struct address_space *mapping = inode->i_mapping;
453
454         BUG_ON(offset & ~HPAGE_MASK);
455         pgoff = offset >> PAGE_SHIFT;
456
457         i_size_write(inode, offset);
458         spin_lock(&mapping->i_mmap_lock);
459         if (!prio_tree_empty(&mapping->i_mmap))
460                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
461         spin_unlock(&mapping->i_mmap_lock);
462         truncate_hugepages(inode, offset);
463         return 0;
464 }
465
466 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
467 {
468         struct inode *inode = dentry->d_inode;
469         int error;
470         unsigned int ia_valid = attr->ia_valid;
471
472         BUG_ON(!inode);
473
474         error = inode_change_ok(inode, attr);
475         if (error)
476                 goto out;
477
478         if (ia_valid & ATTR_SIZE) {
479                 error = -EINVAL;
480                 if (!(attr->ia_size & ~HPAGE_MASK))
481                         error = hugetlb_vmtruncate(inode, attr->ia_size);
482                 if (error)
483                         goto out;
484                 attr->ia_valid &= ~ATTR_SIZE;
485         }
486         error = inode_setattr(inode, attr);
487 out:
488         return error;
489 }
490
491 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid, 
492                                         gid_t gid, int mode, dev_t dev)
493 {
494         struct inode *inode;
495
496         inode = new_inode(sb);
497         if (inode) {
498                 struct hugetlbfs_inode_info *info;
499                 inode->i_mode = mode;
500                 inode->i_uid = uid;
501                 inode->i_gid = gid;
502                 inode->i_blocks = 0;
503                 inode->i_mapping->a_ops = &hugetlbfs_aops;
504                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
505                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
506                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
507                 info = HUGETLBFS_I(inode);
508                 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
509                 switch (mode & S_IFMT) {
510                 default:
511                         init_special_inode(inode, mode, dev);
512                         break;
513                 case S_IFREG:
514                         inode->i_op = &hugetlbfs_inode_operations;
515                         inode->i_fop = &hugetlbfs_file_operations;
516                         break;
517                 case S_IFDIR:
518                         inode->i_op = &hugetlbfs_dir_inode_operations;
519                         inode->i_fop = &simple_dir_operations;
520
521                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
522                         inc_nlink(inode);
523                         break;
524                 case S_IFLNK:
525                         inode->i_op = &page_symlink_inode_operations;
526                         break;
527                 }
528         }
529         return inode;
530 }
531
532 /*
533  * File creation. Allocate an inode, and we're done..
534  */
535 static int hugetlbfs_mknod(struct inode *dir,
536                         struct dentry *dentry, int mode, dev_t dev)
537 {
538         struct inode *inode;
539         int error = -ENOSPC;
540         gid_t gid;
541
542         if (dir->i_mode & S_ISGID) {
543                 gid = dir->i_gid;
544                 if (S_ISDIR(mode))
545                         mode |= S_ISGID;
546         } else {
547                 gid = current->fsgid;
548         }
549         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
550         if (inode) {
551                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
552                 d_instantiate(dentry, inode);
553                 dget(dentry);   /* Extra count - pin the dentry in core */
554                 error = 0;
555         }
556         return error;
557 }
558
559 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
560 {
561         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
562         if (!retval)
563                 inc_nlink(dir);
564         return retval;
565 }
566
567 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
568 {
569         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
570 }
571
572 static int hugetlbfs_symlink(struct inode *dir,
573                         struct dentry *dentry, const char *symname)
574 {
575         struct inode *inode;
576         int error = -ENOSPC;
577         gid_t gid;
578
579         if (dir->i_mode & S_ISGID)
580                 gid = dir->i_gid;
581         else
582                 gid = current->fsgid;
583
584         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
585                                         gid, S_IFLNK|S_IRWXUGO, 0);
586         if (inode) {
587                 int l = strlen(symname)+1;
588                 error = page_symlink(inode, symname, l);
589                 if (!error) {
590                         d_instantiate(dentry, inode);
591                         dget(dentry);
592                 } else
593                         iput(inode);
594         }
595         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
596
597         return error;
598 }
599
600 /*
601  * mark the head page dirty
602  */
603 static int hugetlbfs_set_page_dirty(struct page *page)
604 {
605         struct page *head = compound_head(page);
606
607         SetPageDirty(head);
608         return 0;
609 }
610
611 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
612 {
613         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
614
615         buf->f_type = HUGETLBFS_MAGIC;
616         buf->f_bsize = HPAGE_SIZE;
617         if (sbinfo) {
618                 spin_lock(&sbinfo->stat_lock);
619                 /* If no limits set, just report 0 for max/free/used
620                  * blocks, like simple_statfs() */
621                 if (sbinfo->max_blocks >= 0) {
622                         buf->f_blocks = sbinfo->max_blocks;
623                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
624                         buf->f_files = sbinfo->max_inodes;
625                         buf->f_ffree = sbinfo->free_inodes;
626                 }
627                 spin_unlock(&sbinfo->stat_lock);
628         }
629         buf->f_namelen = NAME_MAX;
630         return 0;
631 }
632
633 static void hugetlbfs_put_super(struct super_block *sb)
634 {
635         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
636
637         if (sbi) {
638                 sb->s_fs_info = NULL;
639                 kfree(sbi);
640         }
641 }
642
643 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
644 {
645         if (sbinfo->free_inodes >= 0) {
646                 spin_lock(&sbinfo->stat_lock);
647                 if (unlikely(!sbinfo->free_inodes)) {
648                         spin_unlock(&sbinfo->stat_lock);
649                         return 0;
650                 }
651                 sbinfo->free_inodes--;
652                 spin_unlock(&sbinfo->stat_lock);
653         }
654
655         return 1;
656 }
657
658 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
659 {
660         if (sbinfo->free_inodes >= 0) {
661                 spin_lock(&sbinfo->stat_lock);
662                 sbinfo->free_inodes++;
663                 spin_unlock(&sbinfo->stat_lock);
664         }
665 }
666
667
668 static struct kmem_cache *hugetlbfs_inode_cachep;
669
670 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
671 {
672         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
673         struct hugetlbfs_inode_info *p;
674
675         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
676                 return NULL;
677         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
678         if (unlikely(!p)) {
679                 hugetlbfs_inc_free_inodes(sbinfo);
680                 return NULL;
681         }
682         return &p->vfs_inode;
683 }
684
685 static void hugetlbfs_destroy_inode(struct inode *inode)
686 {
687         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
688         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
689         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
690 }
691
692 static const struct address_space_operations hugetlbfs_aops = {
693         .readpage       = hugetlbfs_readpage,
694         .write_begin    = hugetlbfs_write_begin,
695         .write_end      = hugetlbfs_write_end,
696         .set_page_dirty = hugetlbfs_set_page_dirty,
697 };
698
699
700 static void init_once(struct kmem_cache *cachep, void *foo)
701 {
702         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
703
704         inode_init_once(&ei->vfs_inode);
705 }
706
707 const struct file_operations hugetlbfs_file_operations = {
708         .read                   = hugetlbfs_read,
709         .mmap                   = hugetlbfs_file_mmap,
710         .fsync                  = simple_sync_file,
711         .get_unmapped_area      = hugetlb_get_unmapped_area,
712 };
713
714 static const struct inode_operations hugetlbfs_dir_inode_operations = {
715         .create         = hugetlbfs_create,
716         .lookup         = simple_lookup,
717         .link           = simple_link,
718         .unlink         = simple_unlink,
719         .symlink        = hugetlbfs_symlink,
720         .mkdir          = hugetlbfs_mkdir,
721         .rmdir          = simple_rmdir,
722         .mknod          = hugetlbfs_mknod,
723         .rename         = simple_rename,
724         .setattr        = hugetlbfs_setattr,
725 };
726
727 static const struct inode_operations hugetlbfs_inode_operations = {
728         .setattr        = hugetlbfs_setattr,
729 };
730
731 static const struct super_operations hugetlbfs_ops = {
732         .alloc_inode    = hugetlbfs_alloc_inode,
733         .destroy_inode  = hugetlbfs_destroy_inode,
734         .statfs         = hugetlbfs_statfs,
735         .delete_inode   = hugetlbfs_delete_inode,
736         .drop_inode     = hugetlbfs_drop_inode,
737         .put_super      = hugetlbfs_put_super,
738 };
739
740 static int
741 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
742 {
743         char *p, *rest;
744         substring_t args[MAX_OPT_ARGS];
745         int option;
746
747         if (!options)
748                 return 0;
749
750         while ((p = strsep(&options, ",")) != NULL) {
751                 int token;
752                 if (!*p)
753                         continue;
754
755                 token = match_token(p, tokens, args);
756                 switch (token) {
757                 case Opt_uid:
758                         if (match_int(&args[0], &option))
759                                 goto bad_val;
760                         pconfig->uid = option;
761                         break;
762
763                 case Opt_gid:
764                         if (match_int(&args[0], &option))
765                                 goto bad_val;
766                         pconfig->gid = option;
767                         break;
768
769                 case Opt_mode:
770                         if (match_octal(&args[0], &option))
771                                 goto bad_val;
772                         pconfig->mode = option & 0777U;
773                         break;
774
775                 case Opt_size: {
776                         unsigned long long size;
777                         /* memparse() will accept a K/M/G without a digit */
778                         if (!isdigit(*args[0].from))
779                                 goto bad_val;
780                         size = memparse(args[0].from, &rest);
781                         if (*rest == '%') {
782                                 size <<= HPAGE_SHIFT;
783                                 size *= max_huge_pages;
784                                 do_div(size, 100);
785                         }
786                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
787                         break;
788                 }
789
790                 case Opt_nr_inodes:
791                         /* memparse() will accept a K/M/G without a digit */
792                         if (!isdigit(*args[0].from))
793                                 goto bad_val;
794                         pconfig->nr_inodes = memparse(args[0].from, &rest);
795                         break;
796
797                 default:
798                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
799                                  p);
800                         return -EINVAL;
801                         break;
802                 }
803         }
804         return 0;
805
806 bad_val:
807         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
808                args[0].from, p);
809         return 1;
810 }
811
812 static int
813 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
814 {
815         struct inode * inode;
816         struct dentry * root;
817         int ret;
818         struct hugetlbfs_config config;
819         struct hugetlbfs_sb_info *sbinfo;
820
821         config.nr_blocks = -1; /* No limit on size by default */
822         config.nr_inodes = -1; /* No limit on number of inodes by default */
823         config.uid = current->fsuid;
824         config.gid = current->fsgid;
825         config.mode = 0755;
826         ret = hugetlbfs_parse_options(data, &config);
827         if (ret)
828                 return ret;
829
830         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
831         if (!sbinfo)
832                 return -ENOMEM;
833         sb->s_fs_info = sbinfo;
834         spin_lock_init(&sbinfo->stat_lock);
835         sbinfo->max_blocks = config.nr_blocks;
836         sbinfo->free_blocks = config.nr_blocks;
837         sbinfo->max_inodes = config.nr_inodes;
838         sbinfo->free_inodes = config.nr_inodes;
839         sb->s_maxbytes = MAX_LFS_FILESIZE;
840         sb->s_blocksize = HPAGE_SIZE;
841         sb->s_blocksize_bits = HPAGE_SHIFT;
842         sb->s_magic = HUGETLBFS_MAGIC;
843         sb->s_op = &hugetlbfs_ops;
844         sb->s_time_gran = 1;
845         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
846                                         S_IFDIR | config.mode, 0);
847         if (!inode)
848                 goto out_free;
849
850         root = d_alloc_root(inode);
851         if (!root) {
852                 iput(inode);
853                 goto out_free;
854         }
855         sb->s_root = root;
856         return 0;
857 out_free:
858         kfree(sbinfo);
859         return -ENOMEM;
860 }
861
862 int hugetlb_get_quota(struct address_space *mapping)
863 {
864         int ret = 0;
865         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
866
867         if (sbinfo->free_blocks > -1) {
868                 spin_lock(&sbinfo->stat_lock);
869                 if (sbinfo->free_blocks > 0)
870                         sbinfo->free_blocks--;
871                 else
872                         ret = -ENOMEM;
873                 spin_unlock(&sbinfo->stat_lock);
874         }
875
876         return ret;
877 }
878
879 void hugetlb_put_quota(struct address_space *mapping)
880 {
881         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
882
883         if (sbinfo->free_blocks > -1) {
884                 spin_lock(&sbinfo->stat_lock);
885                 sbinfo->free_blocks++;
886                 spin_unlock(&sbinfo->stat_lock);
887         }
888 }
889
890 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
891         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
892 {
893         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
894 }
895
896 static struct file_system_type hugetlbfs_fs_type = {
897         .name           = "hugetlbfs",
898         .get_sb         = hugetlbfs_get_sb,
899         .kill_sb        = kill_litter_super,
900 };
901
902 static struct vfsmount *hugetlbfs_vfsmount;
903
904 static int can_do_hugetlb_shm(void)
905 {
906         return likely(capable(CAP_IPC_LOCK) ||
907                         in_group_p(sysctl_hugetlb_shm_group) ||
908                         can_do_mlock());
909 }
910
911 struct file *hugetlb_file_setup(const char *name, size_t size)
912 {
913         int error = -ENOMEM;
914         struct file *file;
915         struct inode *inode;
916         struct dentry *dentry, *root;
917         struct qstr quick_string;
918
919         if (!hugetlbfs_vfsmount)
920                 return ERR_PTR(-ENOENT);
921
922         if (!can_do_hugetlb_shm())
923                 return ERR_PTR(-EPERM);
924
925         if (!user_shm_lock(size, current->user))
926                 return ERR_PTR(-ENOMEM);
927
928         root = hugetlbfs_vfsmount->mnt_root;
929         quick_string.name = name;
930         quick_string.len = strlen(quick_string.name);
931         quick_string.hash = 0;
932         dentry = d_alloc(root, &quick_string);
933         if (!dentry)
934                 goto out_shm_unlock;
935
936         error = -ENOSPC;
937         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
938                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
939         if (!inode)
940                 goto out_dentry;
941
942         error = -ENOMEM;
943         if (hugetlb_reserve_pages(inode, 0, size >> HPAGE_SHIFT))
944                 goto out_inode;
945
946         d_instantiate(dentry, inode);
947         inode->i_size = size;
948         inode->i_nlink = 0;
949
950         error = -ENFILE;
951         file = alloc_file(hugetlbfs_vfsmount, dentry,
952                         FMODE_WRITE | FMODE_READ,
953                         &hugetlbfs_file_operations);
954         if (!file)
955                 goto out_inode;
956
957         return file;
958
959 out_inode:
960         iput(inode);
961 out_dentry:
962         dput(dentry);
963 out_shm_unlock:
964         user_shm_unlock(size, current->user);
965         return ERR_PTR(error);
966 }
967
968 static int __init init_hugetlbfs_fs(void)
969 {
970         int error;
971         struct vfsmount *vfsmount;
972
973         error = bdi_init(&hugetlbfs_backing_dev_info);
974         if (error)
975                 return error;
976
977         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
978                                         sizeof(struct hugetlbfs_inode_info),
979                                         0, 0, init_once);
980         if (hugetlbfs_inode_cachep == NULL)
981                 goto out2;
982
983         error = register_filesystem(&hugetlbfs_fs_type);
984         if (error)
985                 goto out;
986
987         vfsmount = kern_mount(&hugetlbfs_fs_type);
988
989         if (!IS_ERR(vfsmount)) {
990                 hugetlbfs_vfsmount = vfsmount;
991                 return 0;
992         }
993
994         error = PTR_ERR(vfsmount);
995
996  out:
997         if (error)
998                 kmem_cache_destroy(hugetlbfs_inode_cachep);
999  out2:
1000         bdi_destroy(&hugetlbfs_backing_dev_info);
1001         return error;
1002 }
1003
1004 static void __exit exit_hugetlbfs_fs(void)
1005 {
1006         kmem_cache_destroy(hugetlbfs_inode_cachep);
1007         unregister_filesystem(&hugetlbfs_fs_type);
1008         bdi_destroy(&hugetlbfs_backing_dev_info);
1009 }
1010
1011 module_init(init_hugetlbfs_fs)
1012 module_exit(exit_hugetlbfs_fs)
1013
1014 MODULE_LICENSE("GPL");