Untangling ima mess, part 1: alloc_file()
[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/slab.h>
30 #include <linux/dnotify.h>
31 #include <linux/statfs.h>
32 #include <linux/security.h>
33 #include <linux/magic.h>
34
35 #include <asm/uaccess.h>
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         .name           = "hugetlbfs",
45         .ra_pages       = 0,    /* No readahead */
46         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
47 };
48
49 int sysctl_hugetlb_shm_group;
50
51 enum {
52         Opt_size, Opt_nr_inodes,
53         Opt_mode, Opt_uid, Opt_gid,
54         Opt_pagesize,
55         Opt_err,
56 };
57
58 static const match_table_t tokens = {
59         {Opt_size,      "size=%s"},
60         {Opt_nr_inodes, "nr_inodes=%s"},
61         {Opt_mode,      "mode=%o"},
62         {Opt_uid,       "uid=%u"},
63         {Opt_gid,       "gid=%u"},
64         {Opt_pagesize,  "pagesize=%s"},
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         struct hstate *h = hstate_file(file);
84
85         /*
86          * vma address alignment (but not the pgoff alignment) has
87          * already been checked by prepare_hugepage_range.  If you add
88          * any error returns here, do so after setting VM_HUGETLB, so
89          * is_vm_hugetlb_page tests below unmap_region go the right
90          * way when do_mmap_pgoff unwinds (may be important on powerpc
91          * and ia64).
92          */
93         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
94         vma->vm_ops = &hugetlb_vm_ops;
95
96         if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
97                 return -EINVAL;
98
99         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
100
101         mutex_lock(&inode->i_mutex);
102         file_accessed(file);
103
104         ret = -ENOMEM;
105         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
106
107         if (hugetlb_reserve_pages(inode,
108                                 vma->vm_pgoff >> huge_page_order(h),
109                                 len >> huge_page_shift(h), vma,
110                                 vma->vm_flags))
111                 goto out;
112
113         ret = 0;
114         hugetlb_prefault_arch_hook(vma->vm_mm);
115         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
116                 inode->i_size = len;
117 out:
118         mutex_unlock(&inode->i_mutex);
119
120         return ret;
121 }
122
123 /*
124  * Called under down_write(mmap_sem).
125  */
126
127 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
128 static unsigned long
129 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
130                 unsigned long len, unsigned long pgoff, unsigned long flags)
131 {
132         struct mm_struct *mm = current->mm;
133         struct vm_area_struct *vma;
134         unsigned long start_addr;
135         struct hstate *h = hstate_file(file);
136
137         if (len & ~huge_page_mask(h))
138                 return -EINVAL;
139         if (len > TASK_SIZE)
140                 return -ENOMEM;
141
142         if (flags & MAP_FIXED) {
143                 if (prepare_hugepage_range(file, addr, len))
144                         return -EINVAL;
145                 return addr;
146         }
147
148         if (addr) {
149                 addr = ALIGN(addr, huge_page_size(h));
150                 vma = find_vma(mm, addr);
151                 if (TASK_SIZE - len >= addr &&
152                     (!vma || addr + len <= vma->vm_start))
153                         return addr;
154         }
155
156         start_addr = mm->free_area_cache;
157
158         if (len <= mm->cached_hole_size)
159                 start_addr = TASK_UNMAPPED_BASE;
160
161 full_search:
162         addr = ALIGN(start_addr, huge_page_size(h));
163
164         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
165                 /* At this point:  (!vma || addr < vma->vm_end). */
166                 if (TASK_SIZE - len < addr) {
167                         /*
168                          * Start a new search - just in case we missed
169                          * some holes.
170                          */
171                         if (start_addr != TASK_UNMAPPED_BASE) {
172                                 start_addr = TASK_UNMAPPED_BASE;
173                                 goto full_search;
174                         }
175                         return -ENOMEM;
176                 }
177
178                 if (!vma || addr + len <= vma->vm_start)
179                         return addr;
180                 addr = ALIGN(vma->vm_end, huge_page_size(h));
181         }
182 }
183 #endif
184
185 static int
186 hugetlbfs_read_actor(struct page *page, unsigned long offset,
187                         char __user *buf, unsigned long count,
188                         unsigned long size)
189 {
190         char *kaddr;
191         unsigned long left, copied = 0;
192         int i, chunksize;
193
194         if (size > count)
195                 size = count;
196
197         /* Find which 4k chunk and offset with in that chunk */
198         i = offset >> PAGE_CACHE_SHIFT;
199         offset = offset & ~PAGE_CACHE_MASK;
200
201         while (size) {
202                 chunksize = PAGE_CACHE_SIZE;
203                 if (offset)
204                         chunksize -= offset;
205                 if (chunksize > size)
206                         chunksize = size;
207                 kaddr = kmap(&page[i]);
208                 left = __copy_to_user(buf, kaddr + offset, chunksize);
209                 kunmap(&page[i]);
210                 if (left) {
211                         copied += (chunksize - left);
212                         break;
213                 }
214                 offset = 0;
215                 size -= chunksize;
216                 buf += chunksize;
217                 copied += chunksize;
218                 i++;
219         }
220         return copied ? copied : -EFAULT;
221 }
222
223 /*
224  * Support for read() - Find the page attached to f_mapping and copy out the
225  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
226  * since it has PAGE_CACHE_SIZE assumptions.
227  */
228 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
229                               size_t len, loff_t *ppos)
230 {
231         struct hstate *h = hstate_file(filp);
232         struct address_space *mapping = filp->f_mapping;
233         struct inode *inode = mapping->host;
234         unsigned long index = *ppos >> huge_page_shift(h);
235         unsigned long offset = *ppos & ~huge_page_mask(h);
236         unsigned long end_index;
237         loff_t isize;
238         ssize_t retval = 0;
239
240         mutex_lock(&inode->i_mutex);
241
242         /* validate length */
243         if (len == 0)
244                 goto out;
245
246         isize = i_size_read(inode);
247         if (!isize)
248                 goto out;
249
250         end_index = (isize - 1) >> huge_page_shift(h);
251         for (;;) {
252                 struct page *page;
253                 unsigned long nr, ret;
254                 int ra;
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                                 ra = -EFAULT;
278                         else
279                                 ra = 0;
280                 } else {
281                         /*
282                          * We have the page, copy it to user space buffer.
283                          */
284                         ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
285                         ret = ra;
286                 }
287                 if (ra < 0) {
288                         if (retval == 0)
289                                 retval = ra;
290                         if (page)
291                                 page_cache_release(page);
292                         goto out;
293                 }
294
295                 offset += ret;
296                 retval += ret;
297                 len -= ret;
298                 index += offset >> huge_page_shift(h);
299                 offset &= ~huge_page_mask(h);
300
301                 if (page)
302                         page_cache_release(page);
303
304                 /* short read or no more work */
305                 if ((ret != nr) || (len == 0))
306                         break;
307         }
308 out:
309         *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
310         mutex_unlock(&inode->i_mutex);
311         return retval;
312 }
313
314 static int hugetlbfs_write_begin(struct file *file,
315                         struct address_space *mapping,
316                         loff_t pos, unsigned len, unsigned flags,
317                         struct page **pagep, void **fsdata)
318 {
319         return -EINVAL;
320 }
321
322 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
323                         loff_t pos, unsigned len, unsigned copied,
324                         struct page *page, void *fsdata)
325 {
326         BUG();
327         return -EINVAL;
328 }
329
330 static void truncate_huge_page(struct page *page)
331 {
332         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
333         ClearPageUptodate(page);
334         remove_from_page_cache(page);
335         put_page(page);
336 }
337
338 static void truncate_hugepages(struct inode *inode, loff_t lstart)
339 {
340         struct hstate *h = hstate_inode(inode);
341         struct address_space *mapping = &inode->i_data;
342         const pgoff_t start = lstart >> huge_page_shift(h);
343         struct pagevec pvec;
344         pgoff_t next;
345         int i, freed = 0;
346
347         pagevec_init(&pvec, 0);
348         next = start;
349         while (1) {
350                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
351                         if (next == start)
352                                 break;
353                         next = start;
354                         continue;
355                 }
356
357                 for (i = 0; i < pagevec_count(&pvec); ++i) {
358                         struct page *page = pvec.pages[i];
359
360                         lock_page(page);
361                         if (page->index > next)
362                                 next = page->index;
363                         ++next;
364                         truncate_huge_page(page);
365                         unlock_page(page);
366                         freed++;
367                 }
368                 huge_pagevec_release(&pvec);
369         }
370         BUG_ON(!lstart && mapping->nrpages);
371         hugetlb_unreserve_pages(inode, start, freed);
372 }
373
374 static void hugetlbfs_delete_inode(struct inode *inode)
375 {
376         truncate_hugepages(inode, 0);
377         clear_inode(inode);
378 }
379
380 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
381 {
382         if (generic_detach_inode(inode)) {
383                 truncate_hugepages(inode, 0);
384                 clear_inode(inode);
385                 destroy_inode(inode);
386         }
387 }
388
389 static void hugetlbfs_drop_inode(struct inode *inode)
390 {
391         if (!inode->i_nlink)
392                 generic_delete_inode(inode);
393         else
394                 hugetlbfs_forget_inode(inode);
395 }
396
397 static inline void
398 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
399 {
400         struct vm_area_struct *vma;
401         struct prio_tree_iter iter;
402
403         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
404                 unsigned long v_offset;
405
406                 /*
407                  * Can the expression below overflow on 32-bit arches?
408                  * No, because the prio_tree returns us only those vmas
409                  * which overlap the truncated area starting at pgoff,
410                  * and no vma on a 32-bit arch can span beyond the 4GB.
411                  */
412                 if (vma->vm_pgoff < pgoff)
413                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
414                 else
415                         v_offset = 0;
416
417                 __unmap_hugepage_range(vma,
418                                 vma->vm_start + v_offset, vma->vm_end, NULL);
419         }
420 }
421
422 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
423 {
424         pgoff_t pgoff;
425         struct address_space *mapping = inode->i_mapping;
426         struct hstate *h = hstate_inode(inode);
427
428         BUG_ON(offset & ~huge_page_mask(h));
429         pgoff = offset >> PAGE_SHIFT;
430
431         i_size_write(inode, offset);
432         spin_lock(&mapping->i_mmap_lock);
433         if (!prio_tree_empty(&mapping->i_mmap))
434                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
435         spin_unlock(&mapping->i_mmap_lock);
436         truncate_hugepages(inode, offset);
437         return 0;
438 }
439
440 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
441 {
442         struct inode *inode = dentry->d_inode;
443         struct hstate *h = hstate_inode(inode);
444         int error;
445         unsigned int ia_valid = attr->ia_valid;
446
447         BUG_ON(!inode);
448
449         error = inode_change_ok(inode, attr);
450         if (error)
451                 goto out;
452
453         if (ia_valid & ATTR_SIZE) {
454                 error = -EINVAL;
455                 if (!(attr->ia_size & ~huge_page_mask(h)))
456                         error = hugetlb_vmtruncate(inode, attr->ia_size);
457                 if (error)
458                         goto out;
459                 attr->ia_valid &= ~ATTR_SIZE;
460         }
461         error = inode_setattr(inode, attr);
462 out:
463         return error;
464 }
465
466 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid, 
467                                         gid_t gid, int mode, dev_t dev)
468 {
469         struct inode *inode;
470
471         inode = new_inode(sb);
472         if (inode) {
473                 struct hugetlbfs_inode_info *info;
474                 inode->i_mode = mode;
475                 inode->i_uid = uid;
476                 inode->i_gid = gid;
477                 inode->i_mapping->a_ops = &hugetlbfs_aops;
478                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
479                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
480                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
481                 info = HUGETLBFS_I(inode);
482                 /*
483                  * The policy is initialized here even if we are creating a
484                  * private inode because initialization simply creates an
485                  * an empty rb tree and calls spin_lock_init(), later when we
486                  * call mpol_free_shared_policy() it will just return because
487                  * the rb tree will still be empty.
488                  */
489                 mpol_shared_policy_init(&info->policy, NULL);
490                 switch (mode & S_IFMT) {
491                 default:
492                         init_special_inode(inode, mode, dev);
493                         break;
494                 case S_IFREG:
495                         inode->i_op = &hugetlbfs_inode_operations;
496                         inode->i_fop = &hugetlbfs_file_operations;
497                         break;
498                 case S_IFDIR:
499                         inode->i_op = &hugetlbfs_dir_inode_operations;
500                         inode->i_fop = &simple_dir_operations;
501
502                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
503                         inc_nlink(inode);
504                         break;
505                 case S_IFLNK:
506                         inode->i_op = &page_symlink_inode_operations;
507                         break;
508                 }
509         }
510         return inode;
511 }
512
513 /*
514  * File creation. Allocate an inode, and we're done..
515  */
516 static int hugetlbfs_mknod(struct inode *dir,
517                         struct dentry *dentry, int mode, dev_t dev)
518 {
519         struct inode *inode;
520         int error = -ENOSPC;
521         gid_t gid;
522
523         if (dir->i_mode & S_ISGID) {
524                 gid = dir->i_gid;
525                 if (S_ISDIR(mode))
526                         mode |= S_ISGID;
527         } else {
528                 gid = current_fsgid();
529         }
530         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
531         if (inode) {
532                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
533                 d_instantiate(dentry, inode);
534                 dget(dentry);   /* Extra count - pin the dentry in core */
535                 error = 0;
536         }
537         return error;
538 }
539
540 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
541 {
542         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
543         if (!retval)
544                 inc_nlink(dir);
545         return retval;
546 }
547
548 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
549 {
550         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
551 }
552
553 static int hugetlbfs_symlink(struct inode *dir,
554                         struct dentry *dentry, const char *symname)
555 {
556         struct inode *inode;
557         int error = -ENOSPC;
558         gid_t gid;
559
560         if (dir->i_mode & S_ISGID)
561                 gid = dir->i_gid;
562         else
563                 gid = current_fsgid();
564
565         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
566                                         gid, S_IFLNK|S_IRWXUGO, 0);
567         if (inode) {
568                 int l = strlen(symname)+1;
569                 error = page_symlink(inode, symname, l);
570                 if (!error) {
571                         d_instantiate(dentry, inode);
572                         dget(dentry);
573                 } else
574                         iput(inode);
575         }
576         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
577
578         return error;
579 }
580
581 /*
582  * mark the head page dirty
583  */
584 static int hugetlbfs_set_page_dirty(struct page *page)
585 {
586         struct page *head = compound_head(page);
587
588         SetPageDirty(head);
589         return 0;
590 }
591
592 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
593 {
594         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
595         struct hstate *h = hstate_inode(dentry->d_inode);
596
597         buf->f_type = HUGETLBFS_MAGIC;
598         buf->f_bsize = huge_page_size(h);
599         if (sbinfo) {
600                 spin_lock(&sbinfo->stat_lock);
601                 /* If no limits set, just report 0 for max/free/used
602                  * blocks, like simple_statfs() */
603                 if (sbinfo->max_blocks >= 0) {
604                         buf->f_blocks = sbinfo->max_blocks;
605                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
606                         buf->f_files = sbinfo->max_inodes;
607                         buf->f_ffree = sbinfo->free_inodes;
608                 }
609                 spin_unlock(&sbinfo->stat_lock);
610         }
611         buf->f_namelen = NAME_MAX;
612         return 0;
613 }
614
615 static void hugetlbfs_put_super(struct super_block *sb)
616 {
617         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
618
619         if (sbi) {
620                 sb->s_fs_info = NULL;
621                 kfree(sbi);
622         }
623 }
624
625 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
626 {
627         if (sbinfo->free_inodes >= 0) {
628                 spin_lock(&sbinfo->stat_lock);
629                 if (unlikely(!sbinfo->free_inodes)) {
630                         spin_unlock(&sbinfo->stat_lock);
631                         return 0;
632                 }
633                 sbinfo->free_inodes--;
634                 spin_unlock(&sbinfo->stat_lock);
635         }
636
637         return 1;
638 }
639
640 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
641 {
642         if (sbinfo->free_inodes >= 0) {
643                 spin_lock(&sbinfo->stat_lock);
644                 sbinfo->free_inodes++;
645                 spin_unlock(&sbinfo->stat_lock);
646         }
647 }
648
649
650 static struct kmem_cache *hugetlbfs_inode_cachep;
651
652 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
653 {
654         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
655         struct hugetlbfs_inode_info *p;
656
657         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
658                 return NULL;
659         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
660         if (unlikely(!p)) {
661                 hugetlbfs_inc_free_inodes(sbinfo);
662                 return NULL;
663         }
664         return &p->vfs_inode;
665 }
666
667 static void hugetlbfs_destroy_inode(struct inode *inode)
668 {
669         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
670         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
671         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
672 }
673
674 static const struct address_space_operations hugetlbfs_aops = {
675         .write_begin    = hugetlbfs_write_begin,
676         .write_end      = hugetlbfs_write_end,
677         .set_page_dirty = hugetlbfs_set_page_dirty,
678 };
679
680
681 static void init_once(void *foo)
682 {
683         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
684
685         inode_init_once(&ei->vfs_inode);
686 }
687
688 const struct file_operations hugetlbfs_file_operations = {
689         .read                   = hugetlbfs_read,
690         .mmap                   = hugetlbfs_file_mmap,
691         .fsync                  = simple_sync_file,
692         .get_unmapped_area      = hugetlb_get_unmapped_area,
693 };
694
695 static const struct inode_operations hugetlbfs_dir_inode_operations = {
696         .create         = hugetlbfs_create,
697         .lookup         = simple_lookup,
698         .link           = simple_link,
699         .unlink         = simple_unlink,
700         .symlink        = hugetlbfs_symlink,
701         .mkdir          = hugetlbfs_mkdir,
702         .rmdir          = simple_rmdir,
703         .mknod          = hugetlbfs_mknod,
704         .rename         = simple_rename,
705         .setattr        = hugetlbfs_setattr,
706 };
707
708 static const struct inode_operations hugetlbfs_inode_operations = {
709         .setattr        = hugetlbfs_setattr,
710 };
711
712 static const struct super_operations hugetlbfs_ops = {
713         .alloc_inode    = hugetlbfs_alloc_inode,
714         .destroy_inode  = hugetlbfs_destroy_inode,
715         .statfs         = hugetlbfs_statfs,
716         .delete_inode   = hugetlbfs_delete_inode,
717         .drop_inode     = hugetlbfs_drop_inode,
718         .put_super      = hugetlbfs_put_super,
719         .show_options   = generic_show_options,
720 };
721
722 static int
723 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
724 {
725         char *p, *rest;
726         substring_t args[MAX_OPT_ARGS];
727         int option;
728         unsigned long long size = 0;
729         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
730
731         if (!options)
732                 return 0;
733
734         while ((p = strsep(&options, ",")) != NULL) {
735                 int token;
736                 if (!*p)
737                         continue;
738
739                 token = match_token(p, tokens, args);
740                 switch (token) {
741                 case Opt_uid:
742                         if (match_int(&args[0], &option))
743                                 goto bad_val;
744                         pconfig->uid = option;
745                         break;
746
747                 case Opt_gid:
748                         if (match_int(&args[0], &option))
749                                 goto bad_val;
750                         pconfig->gid = option;
751                         break;
752
753                 case Opt_mode:
754                         if (match_octal(&args[0], &option))
755                                 goto bad_val;
756                         pconfig->mode = option & 01777U;
757                         break;
758
759                 case Opt_size: {
760                         /* memparse() will accept a K/M/G without a digit */
761                         if (!isdigit(*args[0].from))
762                                 goto bad_val;
763                         size = memparse(args[0].from, &rest);
764                         setsize = SIZE_STD;
765                         if (*rest == '%')
766                                 setsize = SIZE_PERCENT;
767                         break;
768                 }
769
770                 case Opt_nr_inodes:
771                         /* memparse() will accept a K/M/G without a digit */
772                         if (!isdigit(*args[0].from))
773                                 goto bad_val;
774                         pconfig->nr_inodes = memparse(args[0].from, &rest);
775                         break;
776
777                 case Opt_pagesize: {
778                         unsigned long ps;
779                         ps = memparse(args[0].from, &rest);
780                         pconfig->hstate = size_to_hstate(ps);
781                         if (!pconfig->hstate) {
782                                 printk(KERN_ERR
783                                 "hugetlbfs: Unsupported page size %lu MB\n",
784                                         ps >> 20);
785                                 return -EINVAL;
786                         }
787                         break;
788                 }
789
790                 default:
791                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
792                                  p);
793                         return -EINVAL;
794                         break;
795                 }
796         }
797
798         /* Do size after hstate is set up */
799         if (setsize > NO_SIZE) {
800                 struct hstate *h = pconfig->hstate;
801                 if (setsize == SIZE_PERCENT) {
802                         size <<= huge_page_shift(h);
803                         size *= h->max_huge_pages;
804                         do_div(size, 100);
805                 }
806                 pconfig->nr_blocks = (size >> huge_page_shift(h));
807         }
808
809         return 0;
810
811 bad_val:
812         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
813                args[0].from, p);
814         return -EINVAL;
815 }
816
817 static int
818 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
819 {
820         struct inode * inode;
821         struct dentry * root;
822         int ret;
823         struct hugetlbfs_config config;
824         struct hugetlbfs_sb_info *sbinfo;
825
826         save_mount_options(sb, data);
827
828         config.nr_blocks = -1; /* No limit on size by default */
829         config.nr_inodes = -1; /* No limit on number of inodes by default */
830         config.uid = current_fsuid();
831         config.gid = current_fsgid();
832         config.mode = 0755;
833         config.hstate = &default_hstate;
834         ret = hugetlbfs_parse_options(data, &config);
835         if (ret)
836                 return ret;
837
838         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
839         if (!sbinfo)
840                 return -ENOMEM;
841         sb->s_fs_info = sbinfo;
842         sbinfo->hstate = config.hstate;
843         spin_lock_init(&sbinfo->stat_lock);
844         sbinfo->max_blocks = config.nr_blocks;
845         sbinfo->free_blocks = config.nr_blocks;
846         sbinfo->max_inodes = config.nr_inodes;
847         sbinfo->free_inodes = config.nr_inodes;
848         sb->s_maxbytes = MAX_LFS_FILESIZE;
849         sb->s_blocksize = huge_page_size(config.hstate);
850         sb->s_blocksize_bits = huge_page_shift(config.hstate);
851         sb->s_magic = HUGETLBFS_MAGIC;
852         sb->s_op = &hugetlbfs_ops;
853         sb->s_time_gran = 1;
854         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
855                                         S_IFDIR | config.mode, 0);
856         if (!inode)
857                 goto out_free;
858
859         root = d_alloc_root(inode);
860         if (!root) {
861                 iput(inode);
862                 goto out_free;
863         }
864         sb->s_root = root;
865         return 0;
866 out_free:
867         kfree(sbinfo);
868         return -ENOMEM;
869 }
870
871 int hugetlb_get_quota(struct address_space *mapping, long delta)
872 {
873         int ret = 0;
874         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
875
876         if (sbinfo->free_blocks > -1) {
877                 spin_lock(&sbinfo->stat_lock);
878                 if (sbinfo->free_blocks - delta >= 0)
879                         sbinfo->free_blocks -= delta;
880                 else
881                         ret = -ENOMEM;
882                 spin_unlock(&sbinfo->stat_lock);
883         }
884
885         return ret;
886 }
887
888 void hugetlb_put_quota(struct address_space *mapping, long delta)
889 {
890         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
891
892         if (sbinfo->free_blocks > -1) {
893                 spin_lock(&sbinfo->stat_lock);
894                 sbinfo->free_blocks += delta;
895                 spin_unlock(&sbinfo->stat_lock);
896         }
897 }
898
899 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
900         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
901 {
902         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
903 }
904
905 static struct file_system_type hugetlbfs_fs_type = {
906         .name           = "hugetlbfs",
907         .get_sb         = hugetlbfs_get_sb,
908         .kill_sb        = kill_litter_super,
909 };
910
911 static struct vfsmount *hugetlbfs_vfsmount;
912
913 static int can_do_hugetlb_shm(void)
914 {
915         return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
916 }
917
918 struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag,
919                                 struct user_struct **user, int creat_flags)
920 {
921         int error = -ENOMEM;
922         struct file *file;
923         struct inode *inode;
924         struct path path;
925         struct dentry *root;
926         struct qstr quick_string;
927
928         *user = NULL;
929         if (!hugetlbfs_vfsmount)
930                 return ERR_PTR(-ENOENT);
931
932         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
933                 *user = current_user();
934                 if (user_shm_lock(size, *user)) {
935                         WARN_ONCE(1,
936                           "Using mlock ulimits for SHM_HUGETLB deprecated\n");
937                 } else {
938                         *user = NULL;
939                         return ERR_PTR(-EPERM);
940                 }
941         }
942
943         root = hugetlbfs_vfsmount->mnt_root;
944         quick_string.name = name;
945         quick_string.len = strlen(quick_string.name);
946         quick_string.hash = 0;
947         path.dentry = d_alloc(root, &quick_string);
948         if (!path.dentry)
949                 goto out_shm_unlock;
950
951         path.mnt = mntget(hugetlbfs_vfsmount);
952         error = -ENOSPC;
953         inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
954                                 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
955         if (!inode)
956                 goto out_dentry;
957
958         error = -ENOMEM;
959         if (hugetlb_reserve_pages(inode, 0,
960                         size >> huge_page_shift(hstate_inode(inode)), NULL,
961                         acctflag))
962                 goto out_inode;
963
964         d_instantiate(path.dentry, inode);
965         inode->i_size = size;
966         inode->i_nlink = 0;
967
968         error = -ENFILE;
969         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
970                         &hugetlbfs_file_operations);
971         if (!file)
972                 goto out_dentry; /* inode is already attached */
973
974         return file;
975
976 out_inode:
977         iput(inode);
978 out_dentry:
979         path_put(&path);
980 out_shm_unlock:
981         if (*user) {
982                 user_shm_unlock(size, *user);
983                 *user = NULL;
984         }
985         return ERR_PTR(error);
986 }
987
988 static int __init init_hugetlbfs_fs(void)
989 {
990         int error;
991         struct vfsmount *vfsmount;
992
993         error = bdi_init(&hugetlbfs_backing_dev_info);
994         if (error)
995                 return error;
996
997         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
998                                         sizeof(struct hugetlbfs_inode_info),
999                                         0, 0, init_once);
1000         if (hugetlbfs_inode_cachep == NULL)
1001                 goto out2;
1002
1003         error = register_filesystem(&hugetlbfs_fs_type);
1004         if (error)
1005                 goto out;
1006
1007         vfsmount = kern_mount(&hugetlbfs_fs_type);
1008
1009         if (!IS_ERR(vfsmount)) {
1010                 hugetlbfs_vfsmount = vfsmount;
1011                 return 0;
1012         }
1013
1014         error = PTR_ERR(vfsmount);
1015
1016  out:
1017         if (error)
1018                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1019  out2:
1020         bdi_destroy(&hugetlbfs_backing_dev_info);
1021         return error;
1022 }
1023
1024 static void __exit exit_hugetlbfs_fs(void)
1025 {
1026         kmem_cache_destroy(hugetlbfs_inode_cachep);
1027         unregister_filesystem(&hugetlbfs_fs_type);
1028         bdi_destroy(&hugetlbfs_backing_dev_info);
1029 }
1030
1031 module_init(init_hugetlbfs_fs)
1032 module_exit(exit_hugetlbfs_fs)
1033
1034 MODULE_LICENSE("GPL");