Merge branch 'hwpoison' of git://git.kernel.org/pub/scm/linux/kernel/git/ak/linux...
[~shefty/rdma-dev.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 #include <linux/migrate.h>
35
36 #include <asm/uaccess.h>
37
38 static const struct super_operations hugetlbfs_ops;
39 static const struct address_space_operations hugetlbfs_aops;
40 const struct file_operations hugetlbfs_file_operations;
41 static const struct inode_operations hugetlbfs_dir_inode_operations;
42 static const struct inode_operations hugetlbfs_inode_operations;
43
44 static struct backing_dev_info hugetlbfs_backing_dev_info = {
45         .name           = "hugetlbfs",
46         .ra_pages       = 0,    /* No readahead */
47         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
48 };
49
50 int sysctl_hugetlb_shm_group;
51
52 enum {
53         Opt_size, Opt_nr_inodes,
54         Opt_mode, Opt_uid, Opt_gid,
55         Opt_pagesize,
56         Opt_err,
57 };
58
59 static const 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_pagesize,  "pagesize=%s"},
66         {Opt_err,       NULL},
67 };
68
69 static void huge_pagevec_release(struct pagevec *pvec)
70 {
71         int i;
72
73         for (i = 0; i < pagevec_count(pvec); ++i)
74                 put_page(pvec->pages[i]);
75
76         pagevec_reinit(pvec);
77 }
78
79 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
80 {
81         struct inode *inode = file->f_path.dentry->d_inode;
82         loff_t len, vma_len;
83         int ret;
84         struct hstate *h = hstate_file(file);
85
86         /*
87          * vma address alignment (but not the pgoff alignment) has
88          * already been checked by prepare_hugepage_range.  If you add
89          * any error returns here, do so after setting VM_HUGETLB, so
90          * is_vm_hugetlb_page tests below unmap_region go the right
91          * way when do_mmap_pgoff unwinds (may be important on powerpc
92          * and ia64).
93          */
94         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
95         vma->vm_ops = &hugetlb_vm_ops;
96
97         if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
98                 return -EINVAL;
99
100         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
101
102         mutex_lock(&inode->i_mutex);
103         file_accessed(file);
104
105         ret = -ENOMEM;
106         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
107
108         if (hugetlb_reserve_pages(inode,
109                                 vma->vm_pgoff >> huge_page_order(h),
110                                 len >> huge_page_shift(h), vma,
111                                 vma->vm_flags))
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                 int ra;
256
257                 /* nr is the maximum number of bytes to copy from this page */
258                 nr = huge_page_size(h);
259                 if (index >= end_index) {
260                         if (index > end_index)
261                                 goto out;
262                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
263                         if (nr <= offset) {
264                                 goto out;
265                         }
266                 }
267                 nr = nr - offset;
268
269                 /* Find the page */
270                 page = find_get_page(mapping, index);
271                 if (unlikely(page == NULL)) {
272                         /*
273                          * We have a HOLE, zero out the user-buffer for the
274                          * length of the hole or request.
275                          */
276                         ret = len < nr ? len : nr;
277                         if (clear_user(buf, ret))
278                                 ra = -EFAULT;
279                         else
280                                 ra = 0;
281                 } else {
282                         /*
283                          * We have the page, copy it to user space buffer.
284                          */
285                         ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
286                         ret = ra;
287                 }
288                 if (ra < 0) {
289                         if (retval == 0)
290                                 retval = ra;
291                         if (page)
292                                 page_cache_release(page);
293                         goto out;
294                 }
295
296                 offset += ret;
297                 retval += ret;
298                 len -= ret;
299                 index += offset >> huge_page_shift(h);
300                 offset &= ~huge_page_mask(h);
301
302                 if (page)
303                         page_cache_release(page);
304
305                 /* short read or no more work */
306                 if ((ret != nr) || (len == 0))
307                         break;
308         }
309 out:
310         *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
311         mutex_unlock(&inode->i_mutex);
312         return retval;
313 }
314
315 static int hugetlbfs_write_begin(struct file *file,
316                         struct address_space *mapping,
317                         loff_t pos, unsigned len, unsigned flags,
318                         struct page **pagep, void **fsdata)
319 {
320         return -EINVAL;
321 }
322
323 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
324                         loff_t pos, unsigned len, unsigned copied,
325                         struct page *page, void *fsdata)
326 {
327         BUG();
328         return -EINVAL;
329 }
330
331 static void truncate_huge_page(struct page *page)
332 {
333         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
334         ClearPageUptodate(page);
335         remove_from_page_cache(page);
336         put_page(page);
337 }
338
339 static void truncate_hugepages(struct inode *inode, loff_t lstart)
340 {
341         struct hstate *h = hstate_inode(inode);
342         struct address_space *mapping = &inode->i_data;
343         const pgoff_t start = lstart >> huge_page_shift(h);
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                         freed++;
368                 }
369                 huge_pagevec_release(&pvec);
370         }
371         BUG_ON(!lstart && mapping->nrpages);
372         hugetlb_unreserve_pages(inode, start, freed);
373 }
374
375 static void hugetlbfs_evict_inode(struct inode *inode)
376 {
377         truncate_hugepages(inode, 0);
378         end_writeback(inode);
379 }
380
381 static inline void
382 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
383 {
384         struct vm_area_struct *vma;
385         struct prio_tree_iter iter;
386
387         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
388                 unsigned long v_offset;
389
390                 /*
391                  * Can the expression below overflow on 32-bit arches?
392                  * No, because the prio_tree returns us only those vmas
393                  * which overlap the truncated area starting at pgoff,
394                  * and no vma on a 32-bit arch can span beyond the 4GB.
395                  */
396                 if (vma->vm_pgoff < pgoff)
397                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
398                 else
399                         v_offset = 0;
400
401                 __unmap_hugepage_range(vma,
402                                 vma->vm_start + v_offset, vma->vm_end, NULL);
403         }
404 }
405
406 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
407 {
408         pgoff_t pgoff;
409         struct address_space *mapping = inode->i_mapping;
410         struct hstate *h = hstate_inode(inode);
411
412         BUG_ON(offset & ~huge_page_mask(h));
413         pgoff = offset >> PAGE_SHIFT;
414
415         i_size_write(inode, offset);
416         spin_lock(&mapping->i_mmap_lock);
417         if (!prio_tree_empty(&mapping->i_mmap))
418                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
419         spin_unlock(&mapping->i_mmap_lock);
420         truncate_hugepages(inode, offset);
421         return 0;
422 }
423
424 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
425 {
426         struct inode *inode = dentry->d_inode;
427         struct hstate *h = hstate_inode(inode);
428         int error;
429         unsigned int ia_valid = attr->ia_valid;
430
431         BUG_ON(!inode);
432
433         error = inode_change_ok(inode, attr);
434         if (error)
435                 return error;
436
437         if (ia_valid & ATTR_SIZE) {
438                 error = -EINVAL;
439                 if (attr->ia_size & ~huge_page_mask(h))
440                         return -EINVAL;
441                 error = hugetlb_vmtruncate(inode, attr->ia_size);
442                 if (error)
443                         return error;
444         }
445
446         setattr_copy(inode, attr);
447         mark_inode_dirty(inode);
448         return 0;
449 }
450
451 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid, 
452                                         gid_t gid, int mode, dev_t dev)
453 {
454         struct inode *inode;
455
456         inode = new_inode(sb);
457         if (inode) {
458                 struct hugetlbfs_inode_info *info;
459                 inode->i_mode = mode;
460                 inode->i_uid = uid;
461                 inode->i_gid = gid;
462                 inode->i_mapping->a_ops = &hugetlbfs_aops;
463                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
464                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
465                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
466                 info = HUGETLBFS_I(inode);
467                 /*
468                  * The policy is initialized here even if we are creating a
469                  * private inode because initialization simply creates an
470                  * an empty rb tree and calls spin_lock_init(), later when we
471                  * call mpol_free_shared_policy() it will just return because
472                  * the rb tree will still be empty.
473                  */
474                 mpol_shared_policy_init(&info->policy, NULL);
475                 switch (mode & S_IFMT) {
476                 default:
477                         init_special_inode(inode, mode, dev);
478                         break;
479                 case S_IFREG:
480                         inode->i_op = &hugetlbfs_inode_operations;
481                         inode->i_fop = &hugetlbfs_file_operations;
482                         break;
483                 case S_IFDIR:
484                         inode->i_op = &hugetlbfs_dir_inode_operations;
485                         inode->i_fop = &simple_dir_operations;
486
487                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
488                         inc_nlink(inode);
489                         break;
490                 case S_IFLNK:
491                         inode->i_op = &page_symlink_inode_operations;
492                         break;
493                 }
494         }
495         return inode;
496 }
497
498 /*
499  * File creation. Allocate an inode, and we're done..
500  */
501 static int hugetlbfs_mknod(struct inode *dir,
502                         struct dentry *dentry, int mode, dev_t dev)
503 {
504         struct inode *inode;
505         int error = -ENOSPC;
506         gid_t gid;
507
508         if (dir->i_mode & S_ISGID) {
509                 gid = dir->i_gid;
510                 if (S_ISDIR(mode))
511                         mode |= S_ISGID;
512         } else {
513                 gid = current_fsgid();
514         }
515         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
516         if (inode) {
517                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
518                 d_instantiate(dentry, inode);
519                 dget(dentry);   /* Extra count - pin the dentry in core */
520                 error = 0;
521         }
522         return error;
523 }
524
525 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
526 {
527         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
528         if (!retval)
529                 inc_nlink(dir);
530         return retval;
531 }
532
533 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
534 {
535         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
536 }
537
538 static int hugetlbfs_symlink(struct inode *dir,
539                         struct dentry *dentry, const char *symname)
540 {
541         struct inode *inode;
542         int error = -ENOSPC;
543         gid_t gid;
544
545         if (dir->i_mode & S_ISGID)
546                 gid = dir->i_gid;
547         else
548                 gid = current_fsgid();
549
550         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
551                                         gid, S_IFLNK|S_IRWXUGO, 0);
552         if (inode) {
553                 int l = strlen(symname)+1;
554                 error = page_symlink(inode, symname, l);
555                 if (!error) {
556                         d_instantiate(dentry, inode);
557                         dget(dentry);
558                 } else
559                         iput(inode);
560         }
561         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
562
563         return error;
564 }
565
566 /*
567  * mark the head page dirty
568  */
569 static int hugetlbfs_set_page_dirty(struct page *page)
570 {
571         struct page *head = compound_head(page);
572
573         SetPageDirty(head);
574         return 0;
575 }
576
577 static int hugetlbfs_migrate_page(struct address_space *mapping,
578                                 struct page *newpage, struct page *page)
579 {
580         int rc;
581
582         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
583         if (rc)
584                 return rc;
585         migrate_page_copy(newpage, page);
586
587         return 0;
588 }
589
590 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
591 {
592         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
593         struct hstate *h = hstate_inode(dentry->d_inode);
594
595         buf->f_type = HUGETLBFS_MAGIC;
596         buf->f_bsize = huge_page_size(h);
597         if (sbinfo) {
598                 spin_lock(&sbinfo->stat_lock);
599                 /* If no limits set, just report 0 for max/free/used
600                  * blocks, like simple_statfs() */
601                 if (sbinfo->max_blocks >= 0) {
602                         buf->f_blocks = sbinfo->max_blocks;
603                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
604                         buf->f_files = sbinfo->max_inodes;
605                         buf->f_ffree = sbinfo->free_inodes;
606                 }
607                 spin_unlock(&sbinfo->stat_lock);
608         }
609         buf->f_namelen = NAME_MAX;
610         return 0;
611 }
612
613 static void hugetlbfs_put_super(struct super_block *sb)
614 {
615         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
616
617         if (sbi) {
618                 sb->s_fs_info = NULL;
619                 kfree(sbi);
620         }
621 }
622
623 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
624 {
625         if (sbinfo->free_inodes >= 0) {
626                 spin_lock(&sbinfo->stat_lock);
627                 if (unlikely(!sbinfo->free_inodes)) {
628                         spin_unlock(&sbinfo->stat_lock);
629                         return 0;
630                 }
631                 sbinfo->free_inodes--;
632                 spin_unlock(&sbinfo->stat_lock);
633         }
634
635         return 1;
636 }
637
638 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
639 {
640         if (sbinfo->free_inodes >= 0) {
641                 spin_lock(&sbinfo->stat_lock);
642                 sbinfo->free_inodes++;
643                 spin_unlock(&sbinfo->stat_lock);
644         }
645 }
646
647
648 static struct kmem_cache *hugetlbfs_inode_cachep;
649
650 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
651 {
652         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
653         struct hugetlbfs_inode_info *p;
654
655         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
656                 return NULL;
657         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
658         if (unlikely(!p)) {
659                 hugetlbfs_inc_free_inodes(sbinfo);
660                 return NULL;
661         }
662         return &p->vfs_inode;
663 }
664
665 static void hugetlbfs_destroy_inode(struct inode *inode)
666 {
667         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
668         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
669         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
670 }
671
672 static const struct address_space_operations hugetlbfs_aops = {
673         .write_begin    = hugetlbfs_write_begin,
674         .write_end      = hugetlbfs_write_end,
675         .set_page_dirty = hugetlbfs_set_page_dirty,
676         .migratepage    = hugetlbfs_migrate_page,
677 };
678
679
680 static void init_once(void *foo)
681 {
682         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
683
684         inode_init_once(&ei->vfs_inode);
685 }
686
687 const struct file_operations hugetlbfs_file_operations = {
688         .read                   = hugetlbfs_read,
689         .mmap                   = hugetlbfs_file_mmap,
690         .fsync                  = noop_fsync,
691         .get_unmapped_area      = hugetlb_get_unmapped_area,
692         .llseek         = default_llseek,
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         .evict_inode    = hugetlbfs_evict_inode,
716         .statfs         = hugetlbfs_statfs,
717         .put_super      = hugetlbfs_put_super,
718         .show_options   = generic_show_options,
719 };
720
721 static int
722 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
723 {
724         char *p, *rest;
725         substring_t args[MAX_OPT_ARGS];
726         int option;
727         unsigned long long size = 0;
728         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
729
730         if (!options)
731                 return 0;
732
733         while ((p = strsep(&options, ",")) != NULL) {
734                 int token;
735                 if (!*p)
736                         continue;
737
738                 token = match_token(p, tokens, args);
739                 switch (token) {
740                 case Opt_uid:
741                         if (match_int(&args[0], &option))
742                                 goto bad_val;
743                         pconfig->uid = option;
744                         break;
745
746                 case Opt_gid:
747                         if (match_int(&args[0], &option))
748                                 goto bad_val;
749                         pconfig->gid = option;
750                         break;
751
752                 case Opt_mode:
753                         if (match_octal(&args[0], &option))
754                                 goto bad_val;
755                         pconfig->mode = option & 01777U;
756                         break;
757
758                 case Opt_size: {
759                         /* memparse() will accept a K/M/G without a digit */
760                         if (!isdigit(*args[0].from))
761                                 goto bad_val;
762                         size = memparse(args[0].from, &rest);
763                         setsize = SIZE_STD;
764                         if (*rest == '%')
765                                 setsize = SIZE_PERCENT;
766                         break;
767                 }
768
769                 case Opt_nr_inodes:
770                         /* memparse() will accept a K/M/G without a digit */
771                         if (!isdigit(*args[0].from))
772                                 goto bad_val;
773                         pconfig->nr_inodes = memparse(args[0].from, &rest);
774                         break;
775
776                 case Opt_pagesize: {
777                         unsigned long ps;
778                         ps = memparse(args[0].from, &rest);
779                         pconfig->hstate = size_to_hstate(ps);
780                         if (!pconfig->hstate) {
781                                 printk(KERN_ERR
782                                 "hugetlbfs: Unsupported page size %lu MB\n",
783                                         ps >> 20);
784                                 return -EINVAL;
785                         }
786                         break;
787                 }
788
789                 default:
790                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
791                                  p);
792                         return -EINVAL;
793                         break;
794                 }
795         }
796
797         /* Do size after hstate is set up */
798         if (setsize > NO_SIZE) {
799                 struct hstate *h = pconfig->hstate;
800                 if (setsize == SIZE_PERCENT) {
801                         size <<= huge_page_shift(h);
802                         size *= h->max_huge_pages;
803                         do_div(size, 100);
804                 }
805                 pconfig->nr_blocks = (size >> huge_page_shift(h));
806         }
807
808         return 0;
809
810 bad_val:
811         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
812                args[0].from, p);
813         return -EINVAL;
814 }
815
816 static int
817 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
818 {
819         struct inode * inode;
820         struct dentry * root;
821         int ret;
822         struct hugetlbfs_config config;
823         struct hugetlbfs_sb_info *sbinfo;
824
825         save_mount_options(sb, data);
826
827         config.nr_blocks = -1; /* No limit on size by default */
828         config.nr_inodes = -1; /* No limit on number of inodes by default */
829         config.uid = current_fsuid();
830         config.gid = current_fsgid();
831         config.mode = 0755;
832         config.hstate = &default_hstate;
833         ret = hugetlbfs_parse_options(data, &config);
834         if (ret)
835                 return ret;
836
837         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
838         if (!sbinfo)
839                 return -ENOMEM;
840         sb->s_fs_info = sbinfo;
841         sbinfo->hstate = config.hstate;
842         spin_lock_init(&sbinfo->stat_lock);
843         sbinfo->max_blocks = config.nr_blocks;
844         sbinfo->free_blocks = config.nr_blocks;
845         sbinfo->max_inodes = config.nr_inodes;
846         sbinfo->free_inodes = config.nr_inodes;
847         sb->s_maxbytes = MAX_LFS_FILESIZE;
848         sb->s_blocksize = huge_page_size(config.hstate);
849         sb->s_blocksize_bits = huge_page_shift(config.hstate);
850         sb->s_magic = HUGETLBFS_MAGIC;
851         sb->s_op = &hugetlbfs_ops;
852         sb->s_time_gran = 1;
853         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
854                                         S_IFDIR | config.mode, 0);
855         if (!inode)
856                 goto out_free;
857
858         root = d_alloc_root(inode);
859         if (!root) {
860                 iput(inode);
861                 goto out_free;
862         }
863         sb->s_root = root;
864         return 0;
865 out_free:
866         kfree(sbinfo);
867         return -ENOMEM;
868 }
869
870 int hugetlb_get_quota(struct address_space *mapping, long delta)
871 {
872         int ret = 0;
873         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
874
875         if (sbinfo->free_blocks > -1) {
876                 spin_lock(&sbinfo->stat_lock);
877                 if (sbinfo->free_blocks - delta >= 0)
878                         sbinfo->free_blocks -= delta;
879                 else
880                         ret = -ENOMEM;
881                 spin_unlock(&sbinfo->stat_lock);
882         }
883
884         return ret;
885 }
886
887 void hugetlb_put_quota(struct address_space *mapping, long delta)
888 {
889         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
890
891         if (sbinfo->free_blocks > -1) {
892                 spin_lock(&sbinfo->stat_lock);
893                 sbinfo->free_blocks += delta;
894                 spin_unlock(&sbinfo->stat_lock);
895         }
896 }
897
898 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
899         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
900 {
901         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
902 }
903
904 static struct file_system_type hugetlbfs_fs_type = {
905         .name           = "hugetlbfs",
906         .get_sb         = hugetlbfs_get_sb,
907         .kill_sb        = kill_litter_super,
908 };
909
910 static struct vfsmount *hugetlbfs_vfsmount;
911
912 static int can_do_hugetlb_shm(void)
913 {
914         return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
915 }
916
917 struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag,
918                                 struct user_struct **user, int creat_flags)
919 {
920         int error = -ENOMEM;
921         struct file *file;
922         struct inode *inode;
923         struct path path;
924         struct dentry *root;
925         struct qstr quick_string;
926
927         *user = NULL;
928         if (!hugetlbfs_vfsmount)
929                 return ERR_PTR(-ENOENT);
930
931         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
932                 *user = current_user();
933                 if (user_shm_lock(size, *user)) {
934                         WARN_ONCE(1,
935                           "Using mlock ulimits for SHM_HUGETLB deprecated\n");
936                 } else {
937                         *user = NULL;
938                         return ERR_PTR(-EPERM);
939                 }
940         }
941
942         root = hugetlbfs_vfsmount->mnt_root;
943         quick_string.name = name;
944         quick_string.len = strlen(quick_string.name);
945         quick_string.hash = 0;
946         path.dentry = d_alloc(root, &quick_string);
947         if (!path.dentry)
948                 goto out_shm_unlock;
949
950         path.mnt = mntget(hugetlbfs_vfsmount);
951         error = -ENOSPC;
952         inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
953                                 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
954         if (!inode)
955                 goto out_dentry;
956
957         error = -ENOMEM;
958         if (hugetlb_reserve_pages(inode, 0,
959                         size >> huge_page_shift(hstate_inode(inode)), NULL,
960                         acctflag))
961                 goto out_inode;
962
963         d_instantiate(path.dentry, inode);
964         inode->i_size = size;
965         inode->i_nlink = 0;
966
967         error = -ENFILE;
968         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
969                         &hugetlbfs_file_operations);
970         if (!file)
971                 goto out_dentry; /* inode is already attached */
972
973         return file;
974
975 out_inode:
976         iput(inode);
977 out_dentry:
978         path_put(&path);
979 out_shm_unlock:
980         if (*user) {
981                 user_shm_unlock(size, *user);
982                 *user = NULL;
983         }
984         return ERR_PTR(error);
985 }
986
987 static int __init init_hugetlbfs_fs(void)
988 {
989         int error;
990         struct vfsmount *vfsmount;
991
992         error = bdi_init(&hugetlbfs_backing_dev_info);
993         if (error)
994                 return error;
995
996         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
997                                         sizeof(struct hugetlbfs_inode_info),
998                                         0, 0, init_once);
999         if (hugetlbfs_inode_cachep == NULL)
1000                 goto out2;
1001
1002         error = register_filesystem(&hugetlbfs_fs_type);
1003         if (error)
1004                 goto out;
1005
1006         vfsmount = kern_mount(&hugetlbfs_fs_type);
1007
1008         if (!IS_ERR(vfsmount)) {
1009                 hugetlbfs_vfsmount = vfsmount;
1010                 return 0;
1011         }
1012
1013         error = PTR_ERR(vfsmount);
1014
1015  out:
1016         if (error)
1017                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1018  out2:
1019         bdi_destroy(&hugetlbfs_backing_dev_info);
1020         return error;
1021 }
1022
1023 static void __exit exit_hugetlbfs_fs(void)
1024 {
1025         kmem_cache_destroy(hugetlbfs_inode_cachep);
1026         unregister_filesystem(&hugetlbfs_fs_type);
1027         bdi_destroy(&hugetlbfs_backing_dev_info);
1028 }
1029
1030 module_init(init_hugetlbfs_fs)
1031 module_exit(exit_hugetlbfs_fs)
1032
1033 MODULE_LICENSE("GPL");