Merge branch 'for-next' of git://git.o-hand.com/linux-mfd
[linux-2.6] / 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_AND_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_pagesize,
57         Opt_err,
58 };
59
60 static const match_table_t tokens = {
61         {Opt_size,      "size=%s"},
62         {Opt_nr_inodes, "nr_inodes=%s"},
63         {Opt_mode,      "mode=%o"},
64         {Opt_uid,       "uid=%u"},
65         {Opt_gid,       "gid=%u"},
66         {Opt_pagesize,  "pagesize=%s"},
67         {Opt_err,       NULL},
68 };
69
70 static void huge_pagevec_release(struct pagevec *pvec)
71 {
72         int i;
73
74         for (i = 0; i < pagevec_count(pvec); ++i)
75                 put_page(pvec->pages[i]);
76
77         pagevec_reinit(pvec);
78 }
79
80 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
81 {
82         struct inode *inode = file->f_path.dentry->d_inode;
83         loff_t len, vma_len;
84         int ret;
85         struct hstate *h = hstate_file(file);
86
87         /*
88          * vma address alignment (but not the pgoff alignment) has
89          * already been checked by prepare_hugepage_range.  If you add
90          * any error returns here, do so after setting VM_HUGETLB, so
91          * is_vm_hugetlb_page tests below unmap_region go the right
92          * way when do_mmap_pgoff unwinds (may be important on powerpc
93          * and ia64).
94          */
95         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
96         vma->vm_ops = &hugetlb_vm_ops;
97
98         if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
99                 return -EINVAL;
100
101         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
102
103         mutex_lock(&inode->i_mutex);
104         file_accessed(file);
105
106         ret = -ENOMEM;
107         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
108
109         if (hugetlb_reserve_pages(inode,
110                                 vma->vm_pgoff >> huge_page_order(h),
111                                 len >> huge_page_shift(h), vma))
112                 goto out;
113
114         ret = 0;
115         hugetlb_prefault_arch_hook(vma->vm_mm);
116         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
117                 inode->i_size = len;
118 out:
119         mutex_unlock(&inode->i_mutex);
120
121         return ret;
122 }
123
124 /*
125  * Called under down_write(mmap_sem).
126  */
127
128 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
129 static unsigned long
130 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
131                 unsigned long len, unsigned long pgoff, unsigned long flags)
132 {
133         struct mm_struct *mm = current->mm;
134         struct vm_area_struct *vma;
135         unsigned long start_addr;
136         struct hstate *h = hstate_file(file);
137
138         if (len & ~huge_page_mask(h))
139                 return -EINVAL;
140         if (len > TASK_SIZE)
141                 return -ENOMEM;
142
143         if (flags & MAP_FIXED) {
144                 if (prepare_hugepage_range(file, addr, len))
145                         return -EINVAL;
146                 return addr;
147         }
148
149         if (addr) {
150                 addr = ALIGN(addr, huge_page_size(h));
151                 vma = find_vma(mm, addr);
152                 if (TASK_SIZE - len >= addr &&
153                     (!vma || addr + len <= vma->vm_start))
154                         return addr;
155         }
156
157         start_addr = mm->free_area_cache;
158
159         if (len <= mm->cached_hole_size)
160                 start_addr = TASK_UNMAPPED_BASE;
161
162 full_search:
163         addr = ALIGN(start_addr, huge_page_size(h));
164
165         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
166                 /* At this point:  (!vma || addr < vma->vm_end). */
167                 if (TASK_SIZE - len < addr) {
168                         /*
169                          * Start a new search - just in case we missed
170                          * some holes.
171                          */
172                         if (start_addr != TASK_UNMAPPED_BASE) {
173                                 start_addr = TASK_UNMAPPED_BASE;
174                                 goto full_search;
175                         }
176                         return -ENOMEM;
177                 }
178
179                 if (!vma || addr + len <= vma->vm_start)
180                         return addr;
181                 addr = ALIGN(vma->vm_end, huge_page_size(h));
182         }
183 }
184 #endif
185
186 static int
187 hugetlbfs_read_actor(struct page *page, unsigned long offset,
188                         char __user *buf, unsigned long count,
189                         unsigned long size)
190 {
191         char *kaddr;
192         unsigned long left, copied = 0;
193         int i, chunksize;
194
195         if (size > count)
196                 size = count;
197
198         /* Find which 4k chunk and offset with in that chunk */
199         i = offset >> PAGE_CACHE_SHIFT;
200         offset = offset & ~PAGE_CACHE_MASK;
201
202         while (size) {
203                 chunksize = PAGE_CACHE_SIZE;
204                 if (offset)
205                         chunksize -= offset;
206                 if (chunksize > size)
207                         chunksize = size;
208                 kaddr = kmap(&page[i]);
209                 left = __copy_to_user(buf, kaddr + offset, chunksize);
210                 kunmap(&page[i]);
211                 if (left) {
212                         copied += (chunksize - left);
213                         break;
214                 }
215                 offset = 0;
216                 size -= chunksize;
217                 buf += chunksize;
218                 copied += chunksize;
219                 i++;
220         }
221         return copied ? copied : -EFAULT;
222 }
223
224 /*
225  * Support for read() - Find the page attached to f_mapping and copy out the
226  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
227  * since it has PAGE_CACHE_SIZE assumptions.
228  */
229 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
230                               size_t len, loff_t *ppos)
231 {
232         struct hstate *h = hstate_file(filp);
233         struct address_space *mapping = filp->f_mapping;
234         struct inode *inode = mapping->host;
235         unsigned long index = *ppos >> huge_page_shift(h);
236         unsigned long offset = *ppos & ~huge_page_mask(h);
237         unsigned long end_index;
238         loff_t isize;
239         ssize_t retval = 0;
240
241         mutex_lock(&inode->i_mutex);
242
243         /* validate length */
244         if (len == 0)
245                 goto out;
246
247         isize = i_size_read(inode);
248         if (!isize)
249                 goto out;
250
251         end_index = (isize - 1) >> huge_page_shift(h);
252         for (;;) {
253                 struct page *page;
254                 unsigned long nr, ret;
255
256                 /* nr is the maximum number of bytes to copy from this page */
257                 nr = huge_page_size(h);
258                 if (index >= end_index) {
259                         if (index > end_index)
260                                 goto out;
261                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
262                         if (nr <= offset) {
263                                 goto out;
264                         }
265                 }
266                 nr = nr - offset;
267
268                 /* Find the page */
269                 page = find_get_page(mapping, index);
270                 if (unlikely(page == NULL)) {
271                         /*
272                          * We have a HOLE, zero out the user-buffer for the
273                          * length of the hole or request.
274                          */
275                         ret = len < nr ? len : nr;
276                         if (clear_user(buf, ret))
277                                 ret = -EFAULT;
278                 } else {
279                         /*
280                          * We have the page, copy it to user space buffer.
281                          */
282                         ret = hugetlbfs_read_actor(page, offset, buf, len, nr);
283                 }
284                 if (ret < 0) {
285                         if (retval == 0)
286                                 retval = ret;
287                         if (page)
288                                 page_cache_release(page);
289                         goto out;
290                 }
291
292                 offset += ret;
293                 retval += ret;
294                 len -= ret;
295                 index += offset >> huge_page_shift(h);
296                 offset &= ~huge_page_mask(h);
297
298                 if (page)
299                         page_cache_release(page);
300
301                 /* short read or no more work */
302                 if ((ret != nr) || (len == 0))
303                         break;
304         }
305 out:
306         *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
307         mutex_unlock(&inode->i_mutex);
308         return retval;
309 }
310
311 /*
312  * Read a page. Again trivial. If it didn't already exist
313  * in the page cache, it is zero-filled.
314  */
315 static int hugetlbfs_readpage(struct file *file, struct page * page)
316 {
317         unlock_page(page);
318         return -EINVAL;
319 }
320
321 static int hugetlbfs_write_begin(struct file *file,
322                         struct address_space *mapping,
323                         loff_t pos, unsigned len, unsigned flags,
324                         struct page **pagep, void **fsdata)
325 {
326         return -EINVAL;
327 }
328
329 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
330                         loff_t pos, unsigned len, unsigned copied,
331                         struct page *page, void *fsdata)
332 {
333         BUG();
334         return -EINVAL;
335 }
336
337 static void truncate_huge_page(struct page *page)
338 {
339         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
340         ClearPageUptodate(page);
341         remove_from_page_cache(page);
342         put_page(page);
343 }
344
345 static void truncate_hugepages(struct inode *inode, loff_t lstart)
346 {
347         struct hstate *h = hstate_inode(inode);
348         struct address_space *mapping = &inode->i_data;
349         const pgoff_t start = lstart >> huge_page_shift(h);
350         struct pagevec pvec;
351         pgoff_t next;
352         int i, freed = 0;
353
354         pagevec_init(&pvec, 0);
355         next = start;
356         while (1) {
357                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
358                         if (next == start)
359                                 break;
360                         next = start;
361                         continue;
362                 }
363
364                 for (i = 0; i < pagevec_count(&pvec); ++i) {
365                         struct page *page = pvec.pages[i];
366
367                         lock_page(page);
368                         if (page->index > next)
369                                 next = page->index;
370                         ++next;
371                         truncate_huge_page(page);
372                         unlock_page(page);
373                         freed++;
374                 }
375                 huge_pagevec_release(&pvec);
376         }
377         BUG_ON(!lstart && mapping->nrpages);
378         hugetlb_unreserve_pages(inode, start, freed);
379 }
380
381 static void hugetlbfs_delete_inode(struct inode *inode)
382 {
383         truncate_hugepages(inode, 0);
384         clear_inode(inode);
385 }
386
387 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
388 {
389         struct super_block *sb = inode->i_sb;
390
391         if (!hlist_unhashed(&inode->i_hash)) {
392                 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
393                         list_move(&inode->i_list, &inode_unused);
394                 inodes_stat.nr_unused++;
395                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
396                         spin_unlock(&inode_lock);
397                         return;
398                 }
399                 inode->i_state |= I_WILL_FREE;
400                 spin_unlock(&inode_lock);
401                 /*
402                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
403                  * in our backing_dev_info.
404                  */
405                 write_inode_now(inode, 1);
406                 spin_lock(&inode_lock);
407                 inode->i_state &= ~I_WILL_FREE;
408                 inodes_stat.nr_unused--;
409                 hlist_del_init(&inode->i_hash);
410         }
411         list_del_init(&inode->i_list);
412         list_del_init(&inode->i_sb_list);
413         inode->i_state |= I_FREEING;
414         inodes_stat.nr_inodes--;
415         spin_unlock(&inode_lock);
416         truncate_hugepages(inode, 0);
417         clear_inode(inode);
418         destroy_inode(inode);
419 }
420
421 static void hugetlbfs_drop_inode(struct inode *inode)
422 {
423         if (!inode->i_nlink)
424                 generic_delete_inode(inode);
425         else
426                 hugetlbfs_forget_inode(inode);
427 }
428
429 static inline void
430 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
431 {
432         struct vm_area_struct *vma;
433         struct prio_tree_iter iter;
434
435         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
436                 unsigned long v_offset;
437
438                 /*
439                  * Can the expression below overflow on 32-bit arches?
440                  * No, because the prio_tree returns us only those vmas
441                  * which overlap the truncated area starting at pgoff,
442                  * and no vma on a 32-bit arch can span beyond the 4GB.
443                  */
444                 if (vma->vm_pgoff < pgoff)
445                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
446                 else
447                         v_offset = 0;
448
449                 __unmap_hugepage_range(vma,
450                                 vma->vm_start + v_offset, vma->vm_end, NULL);
451         }
452 }
453
454 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
455 {
456         pgoff_t pgoff;
457         struct address_space *mapping = inode->i_mapping;
458         struct hstate *h = hstate_inode(inode);
459
460         BUG_ON(offset & ~huge_page_mask(h));
461         pgoff = offset >> PAGE_SHIFT;
462
463         i_size_write(inode, offset);
464         spin_lock(&mapping->i_mmap_lock);
465         if (!prio_tree_empty(&mapping->i_mmap))
466                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
467         spin_unlock(&mapping->i_mmap_lock);
468         truncate_hugepages(inode, offset);
469         return 0;
470 }
471
472 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
473 {
474         struct inode *inode = dentry->d_inode;
475         struct hstate *h = hstate_inode(inode);
476         int error;
477         unsigned int ia_valid = attr->ia_valid;
478
479         BUG_ON(!inode);
480
481         error = inode_change_ok(inode, attr);
482         if (error)
483                 goto out;
484
485         if (ia_valid & ATTR_SIZE) {
486                 error = -EINVAL;
487                 if (!(attr->ia_size & ~huge_page_mask(h)))
488                         error = hugetlb_vmtruncate(inode, attr->ia_size);
489                 if (error)
490                         goto out;
491                 attr->ia_valid &= ~ATTR_SIZE;
492         }
493         error = inode_setattr(inode, attr);
494 out:
495         return error;
496 }
497
498 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid, 
499                                         gid_t gid, int mode, dev_t dev)
500 {
501         struct inode *inode;
502
503         inode = new_inode(sb);
504         if (inode) {
505                 struct hugetlbfs_inode_info *info;
506                 inode->i_mode = mode;
507                 inode->i_uid = uid;
508                 inode->i_gid = gid;
509                 inode->i_mapping->a_ops = &hugetlbfs_aops;
510                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
511                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
512                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
513                 info = HUGETLBFS_I(inode);
514                 mpol_shared_policy_init(&info->policy, NULL);
515                 switch (mode & S_IFMT) {
516                 default:
517                         init_special_inode(inode, mode, dev);
518                         break;
519                 case S_IFREG:
520                         inode->i_op = &hugetlbfs_inode_operations;
521                         inode->i_fop = &hugetlbfs_file_operations;
522                         break;
523                 case S_IFDIR:
524                         inode->i_op = &hugetlbfs_dir_inode_operations;
525                         inode->i_fop = &simple_dir_operations;
526
527                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
528                         inc_nlink(inode);
529                         break;
530                 case S_IFLNK:
531                         inode->i_op = &page_symlink_inode_operations;
532                         break;
533                 }
534         }
535         return inode;
536 }
537
538 /*
539  * File creation. Allocate an inode, and we're done..
540  */
541 static int hugetlbfs_mknod(struct inode *dir,
542                         struct dentry *dentry, int mode, dev_t dev)
543 {
544         struct inode *inode;
545         int error = -ENOSPC;
546         gid_t gid;
547
548         if (dir->i_mode & S_ISGID) {
549                 gid = dir->i_gid;
550                 if (S_ISDIR(mode))
551                         mode |= S_ISGID;
552         } else {
553                 gid = current_fsgid();
554         }
555         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
556         if (inode) {
557                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
558                 d_instantiate(dentry, inode);
559                 dget(dentry);   /* Extra count - pin the dentry in core */
560                 error = 0;
561         }
562         return error;
563 }
564
565 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
566 {
567         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
568         if (!retval)
569                 inc_nlink(dir);
570         return retval;
571 }
572
573 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
574 {
575         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
576 }
577
578 static int hugetlbfs_symlink(struct inode *dir,
579                         struct dentry *dentry, const char *symname)
580 {
581         struct inode *inode;
582         int error = -ENOSPC;
583         gid_t gid;
584
585         if (dir->i_mode & S_ISGID)
586                 gid = dir->i_gid;
587         else
588                 gid = current_fsgid();
589
590         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
591                                         gid, S_IFLNK|S_IRWXUGO, 0);
592         if (inode) {
593                 int l = strlen(symname)+1;
594                 error = page_symlink(inode, symname, l);
595                 if (!error) {
596                         d_instantiate(dentry, inode);
597                         dget(dentry);
598                 } else
599                         iput(inode);
600         }
601         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
602
603         return error;
604 }
605
606 /*
607  * mark the head page dirty
608  */
609 static int hugetlbfs_set_page_dirty(struct page *page)
610 {
611         struct page *head = compound_head(page);
612
613         SetPageDirty(head);
614         return 0;
615 }
616
617 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
618 {
619         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
620         struct hstate *h = hstate_inode(dentry->d_inode);
621
622         buf->f_type = HUGETLBFS_MAGIC;
623         buf->f_bsize = huge_page_size(h);
624         if (sbinfo) {
625                 spin_lock(&sbinfo->stat_lock);
626                 /* If no limits set, just report 0 for max/free/used
627                  * blocks, like simple_statfs() */
628                 if (sbinfo->max_blocks >= 0) {
629                         buf->f_blocks = sbinfo->max_blocks;
630                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
631                         buf->f_files = sbinfo->max_inodes;
632                         buf->f_ffree = sbinfo->free_inodes;
633                 }
634                 spin_unlock(&sbinfo->stat_lock);
635         }
636         buf->f_namelen = NAME_MAX;
637         return 0;
638 }
639
640 static void hugetlbfs_put_super(struct super_block *sb)
641 {
642         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
643
644         if (sbi) {
645                 sb->s_fs_info = NULL;
646                 kfree(sbi);
647         }
648 }
649
650 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
651 {
652         if (sbinfo->free_inodes >= 0) {
653                 spin_lock(&sbinfo->stat_lock);
654                 if (unlikely(!sbinfo->free_inodes)) {
655                         spin_unlock(&sbinfo->stat_lock);
656                         return 0;
657                 }
658                 sbinfo->free_inodes--;
659                 spin_unlock(&sbinfo->stat_lock);
660         }
661
662         return 1;
663 }
664
665 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
666 {
667         if (sbinfo->free_inodes >= 0) {
668                 spin_lock(&sbinfo->stat_lock);
669                 sbinfo->free_inodes++;
670                 spin_unlock(&sbinfo->stat_lock);
671         }
672 }
673
674
675 static struct kmem_cache *hugetlbfs_inode_cachep;
676
677 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
678 {
679         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
680         struct hugetlbfs_inode_info *p;
681
682         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
683                 return NULL;
684         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
685         if (unlikely(!p)) {
686                 hugetlbfs_inc_free_inodes(sbinfo);
687                 return NULL;
688         }
689         return &p->vfs_inode;
690 }
691
692 static void hugetlbfs_destroy_inode(struct inode *inode)
693 {
694         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
695         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
696         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
697 }
698
699 static const struct address_space_operations hugetlbfs_aops = {
700         .readpage       = hugetlbfs_readpage,
701         .write_begin    = hugetlbfs_write_begin,
702         .write_end      = hugetlbfs_write_end,
703         .set_page_dirty = hugetlbfs_set_page_dirty,
704 };
705
706
707 static void init_once(void *foo)
708 {
709         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
710
711         inode_init_once(&ei->vfs_inode);
712 }
713
714 const struct file_operations hugetlbfs_file_operations = {
715         .read                   = hugetlbfs_read,
716         .mmap                   = hugetlbfs_file_mmap,
717         .fsync                  = simple_sync_file,
718         .get_unmapped_area      = hugetlb_get_unmapped_area,
719 };
720
721 static const struct inode_operations hugetlbfs_dir_inode_operations = {
722         .create         = hugetlbfs_create,
723         .lookup         = simple_lookup,
724         .link           = simple_link,
725         .unlink         = simple_unlink,
726         .symlink        = hugetlbfs_symlink,
727         .mkdir          = hugetlbfs_mkdir,
728         .rmdir          = simple_rmdir,
729         .mknod          = hugetlbfs_mknod,
730         .rename         = simple_rename,
731         .setattr        = hugetlbfs_setattr,
732 };
733
734 static const struct inode_operations hugetlbfs_inode_operations = {
735         .setattr        = hugetlbfs_setattr,
736 };
737
738 static const struct super_operations hugetlbfs_ops = {
739         .alloc_inode    = hugetlbfs_alloc_inode,
740         .destroy_inode  = hugetlbfs_destroy_inode,
741         .statfs         = hugetlbfs_statfs,
742         .delete_inode   = hugetlbfs_delete_inode,
743         .drop_inode     = hugetlbfs_drop_inode,
744         .put_super      = hugetlbfs_put_super,
745         .show_options   = generic_show_options,
746 };
747
748 static int
749 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
750 {
751         char *p, *rest;
752         substring_t args[MAX_OPT_ARGS];
753         int option;
754         unsigned long long size = 0;
755         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
756
757         if (!options)
758                 return 0;
759
760         while ((p = strsep(&options, ",")) != NULL) {
761                 int token;
762                 if (!*p)
763                         continue;
764
765                 token = match_token(p, tokens, args);
766                 switch (token) {
767                 case Opt_uid:
768                         if (match_int(&args[0], &option))
769                                 goto bad_val;
770                         pconfig->uid = option;
771                         break;
772
773                 case Opt_gid:
774                         if (match_int(&args[0], &option))
775                                 goto bad_val;
776                         pconfig->gid = option;
777                         break;
778
779                 case Opt_mode:
780                         if (match_octal(&args[0], &option))
781                                 goto bad_val;
782                         pconfig->mode = option & 01777U;
783                         break;
784
785                 case Opt_size: {
786                         /* memparse() will accept a K/M/G without a digit */
787                         if (!isdigit(*args[0].from))
788                                 goto bad_val;
789                         size = memparse(args[0].from, &rest);
790                         setsize = SIZE_STD;
791                         if (*rest == '%')
792                                 setsize = SIZE_PERCENT;
793                         break;
794                 }
795
796                 case Opt_nr_inodes:
797                         /* memparse() will accept a K/M/G without a digit */
798                         if (!isdigit(*args[0].from))
799                                 goto bad_val;
800                         pconfig->nr_inodes = memparse(args[0].from, &rest);
801                         break;
802
803                 case Opt_pagesize: {
804                         unsigned long ps;
805                         ps = memparse(args[0].from, &rest);
806                         pconfig->hstate = size_to_hstate(ps);
807                         if (!pconfig->hstate) {
808                                 printk(KERN_ERR
809                                 "hugetlbfs: Unsupported page size %lu MB\n",
810                                         ps >> 20);
811                                 return -EINVAL;
812                         }
813                         break;
814                 }
815
816                 default:
817                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
818                                  p);
819                         return -EINVAL;
820                         break;
821                 }
822         }
823
824         /* Do size after hstate is set up */
825         if (setsize > NO_SIZE) {
826                 struct hstate *h = pconfig->hstate;
827                 if (setsize == SIZE_PERCENT) {
828                         size <<= huge_page_shift(h);
829                         size *= h->max_huge_pages;
830                         do_div(size, 100);
831                 }
832                 pconfig->nr_blocks = (size >> huge_page_shift(h));
833         }
834
835         return 0;
836
837 bad_val:
838         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
839                args[0].from, p);
840         return 1;
841 }
842
843 static int
844 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
845 {
846         struct inode * inode;
847         struct dentry * root;
848         int ret;
849         struct hugetlbfs_config config;
850         struct hugetlbfs_sb_info *sbinfo;
851
852         save_mount_options(sb, data);
853
854         config.nr_blocks = -1; /* No limit on size by default */
855         config.nr_inodes = -1; /* No limit on number of inodes by default */
856         config.uid = current_fsuid();
857         config.gid = current_fsgid();
858         config.mode = 0755;
859         config.hstate = &default_hstate;
860         ret = hugetlbfs_parse_options(data, &config);
861         if (ret)
862                 return ret;
863
864         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
865         if (!sbinfo)
866                 return -ENOMEM;
867         sb->s_fs_info = sbinfo;
868         sbinfo->hstate = config.hstate;
869         spin_lock_init(&sbinfo->stat_lock);
870         sbinfo->max_blocks = config.nr_blocks;
871         sbinfo->free_blocks = config.nr_blocks;
872         sbinfo->max_inodes = config.nr_inodes;
873         sbinfo->free_inodes = config.nr_inodes;
874         sb->s_maxbytes = MAX_LFS_FILESIZE;
875         sb->s_blocksize = huge_page_size(config.hstate);
876         sb->s_blocksize_bits = huge_page_shift(config.hstate);
877         sb->s_magic = HUGETLBFS_MAGIC;
878         sb->s_op = &hugetlbfs_ops;
879         sb->s_time_gran = 1;
880         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
881                                         S_IFDIR | config.mode, 0);
882         if (!inode)
883                 goto out_free;
884
885         root = d_alloc_root(inode);
886         if (!root) {
887                 iput(inode);
888                 goto out_free;
889         }
890         sb->s_root = root;
891         return 0;
892 out_free:
893         kfree(sbinfo);
894         return -ENOMEM;
895 }
896
897 int hugetlb_get_quota(struct address_space *mapping, long delta)
898 {
899         int ret = 0;
900         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
901
902         if (sbinfo->free_blocks > -1) {
903                 spin_lock(&sbinfo->stat_lock);
904                 if (sbinfo->free_blocks - delta >= 0)
905                         sbinfo->free_blocks -= delta;
906                 else
907                         ret = -ENOMEM;
908                 spin_unlock(&sbinfo->stat_lock);
909         }
910
911         return ret;
912 }
913
914 void hugetlb_put_quota(struct address_space *mapping, long delta)
915 {
916         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
917
918         if (sbinfo->free_blocks > -1) {
919                 spin_lock(&sbinfo->stat_lock);
920                 sbinfo->free_blocks += delta;
921                 spin_unlock(&sbinfo->stat_lock);
922         }
923 }
924
925 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
926         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
927 {
928         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
929 }
930
931 static struct file_system_type hugetlbfs_fs_type = {
932         .name           = "hugetlbfs",
933         .get_sb         = hugetlbfs_get_sb,
934         .kill_sb        = kill_litter_super,
935 };
936
937 static struct vfsmount *hugetlbfs_vfsmount;
938
939 static int can_do_hugetlb_shm(void)
940 {
941         return likely(capable(CAP_IPC_LOCK) ||
942                         in_group_p(sysctl_hugetlb_shm_group) ||
943                         can_do_mlock());
944 }
945
946 struct file *hugetlb_file_setup(const char *name, size_t size)
947 {
948         int error = -ENOMEM;
949         struct file *file;
950         struct inode *inode;
951         struct dentry *dentry, *root;
952         struct qstr quick_string;
953         struct user_struct *user = current_user();
954
955         if (!hugetlbfs_vfsmount)
956                 return ERR_PTR(-ENOENT);
957
958         if (!can_do_hugetlb_shm())
959                 return ERR_PTR(-EPERM);
960
961         if (!user_shm_lock(size, user))
962                 return ERR_PTR(-ENOMEM);
963
964         root = hugetlbfs_vfsmount->mnt_root;
965         quick_string.name = name;
966         quick_string.len = strlen(quick_string.name);
967         quick_string.hash = 0;
968         dentry = d_alloc(root, &quick_string);
969         if (!dentry)
970                 goto out_shm_unlock;
971
972         error = -ENOSPC;
973         inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
974                                 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
975         if (!inode)
976                 goto out_dentry;
977
978         error = -ENOMEM;
979         if (hugetlb_reserve_pages(inode, 0,
980                         size >> huge_page_shift(hstate_inode(inode)), NULL))
981                 goto out_inode;
982
983         d_instantiate(dentry, inode);
984         inode->i_size = size;
985         inode->i_nlink = 0;
986
987         error = -ENFILE;
988         file = alloc_file(hugetlbfs_vfsmount, dentry,
989                         FMODE_WRITE | FMODE_READ,
990                         &hugetlbfs_file_operations);
991         if (!file)
992                 goto out_dentry; /* inode is already attached */
993
994         return file;
995
996 out_inode:
997         iput(inode);
998 out_dentry:
999         dput(dentry);
1000 out_shm_unlock:
1001         user_shm_unlock(size, user);
1002         return ERR_PTR(error);
1003 }
1004
1005 static int __init init_hugetlbfs_fs(void)
1006 {
1007         int error;
1008         struct vfsmount *vfsmount;
1009
1010         error = bdi_init(&hugetlbfs_backing_dev_info);
1011         if (error)
1012                 return error;
1013
1014         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1015                                         sizeof(struct hugetlbfs_inode_info),
1016                                         0, 0, init_once);
1017         if (hugetlbfs_inode_cachep == NULL)
1018                 goto out2;
1019
1020         error = register_filesystem(&hugetlbfs_fs_type);
1021         if (error)
1022                 goto out;
1023
1024         vfsmount = kern_mount(&hugetlbfs_fs_type);
1025
1026         if (!IS_ERR(vfsmount)) {
1027                 hugetlbfs_vfsmount = vfsmount;
1028                 return 0;
1029         }
1030
1031         error = PTR_ERR(vfsmount);
1032
1033  out:
1034         if (error)
1035                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1036  out2:
1037         bdi_destroy(&hugetlbfs_backing_dev_info);
1038         return error;
1039 }
1040
1041 static void __exit exit_hugetlbfs_fs(void)
1042 {
1043         kmem_cache_destroy(hugetlbfs_inode_cachep);
1044         unregister_filesystem(&hugetlbfs_fs_type);
1045         bdi_destroy(&hugetlbfs_backing_dev_info);
1046 }
1047
1048 module_init(init_hugetlbfs_fs)
1049 module_exit(exit_hugetlbfs_fs)
1050
1051 MODULE_LICENSE("GPL");