mm: support more pagesizes for MAP_HUGETLB/SHM_HUGETLB
[linux-3.10.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 struct hugetlbfs_config {
45         kuid_t   uid;
46         kgid_t   gid;
47         umode_t mode;
48         long    nr_blocks;
49         long    nr_inodes;
50         struct hstate *hstate;
51 };
52
53 struct hugetlbfs_inode_info {
54         struct shared_policy policy;
55         struct inode vfs_inode;
56 };
57
58 static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
59 {
60         return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
61 }
62
63 static struct backing_dev_info hugetlbfs_backing_dev_info = {
64         .name           = "hugetlbfs",
65         .ra_pages       = 0,    /* No readahead */
66         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
67 };
68
69 int sysctl_hugetlb_shm_group;
70
71 enum {
72         Opt_size, Opt_nr_inodes,
73         Opt_mode, Opt_uid, Opt_gid,
74         Opt_pagesize,
75         Opt_err,
76 };
77
78 static const match_table_t tokens = {
79         {Opt_size,      "size=%s"},
80         {Opt_nr_inodes, "nr_inodes=%s"},
81         {Opt_mode,      "mode=%o"},
82         {Opt_uid,       "uid=%u"},
83         {Opt_gid,       "gid=%u"},
84         {Opt_pagesize,  "pagesize=%s"},
85         {Opt_err,       NULL},
86 };
87
88 static void huge_pagevec_release(struct pagevec *pvec)
89 {
90         int i;
91
92         for (i = 0; i < pagevec_count(pvec); ++i)
93                 put_page(pvec->pages[i]);
94
95         pagevec_reinit(pvec);
96 }
97
98 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
99 {
100         struct inode *inode = file->f_path.dentry->d_inode;
101         loff_t len, vma_len;
102         int ret;
103         struct hstate *h = hstate_file(file);
104
105         /*
106          * vma address alignment (but not the pgoff alignment) has
107          * already been checked by prepare_hugepage_range.  If you add
108          * any error returns here, do so after setting VM_HUGETLB, so
109          * is_vm_hugetlb_page tests below unmap_region go the right
110          * way when do_mmap_pgoff unwinds (may be important on powerpc
111          * and ia64).
112          */
113         vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND | VM_DONTDUMP;
114         vma->vm_ops = &hugetlb_vm_ops;
115
116         if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
117                 return -EINVAL;
118
119         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
120
121         mutex_lock(&inode->i_mutex);
122         file_accessed(file);
123
124         ret = -ENOMEM;
125         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
126
127         if (hugetlb_reserve_pages(inode,
128                                 vma->vm_pgoff >> huge_page_order(h),
129                                 len >> huge_page_shift(h), vma,
130                                 vma->vm_flags))
131                 goto out;
132
133         ret = 0;
134         hugetlb_prefault_arch_hook(vma->vm_mm);
135         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
136                 inode->i_size = len;
137 out:
138         mutex_unlock(&inode->i_mutex);
139
140         return ret;
141 }
142
143 /*
144  * Called under down_write(mmap_sem).
145  */
146
147 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
148 static unsigned long
149 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
150                 unsigned long len, unsigned long pgoff, unsigned long flags)
151 {
152         struct mm_struct *mm = current->mm;
153         struct vm_area_struct *vma;
154         unsigned long start_addr;
155         struct hstate *h = hstate_file(file);
156
157         if (len & ~huge_page_mask(h))
158                 return -EINVAL;
159         if (len > TASK_SIZE)
160                 return -ENOMEM;
161
162         if (flags & MAP_FIXED) {
163                 if (prepare_hugepage_range(file, addr, len))
164                         return -EINVAL;
165                 return addr;
166         }
167
168         if (addr) {
169                 addr = ALIGN(addr, huge_page_size(h));
170                 vma = find_vma(mm, addr);
171                 if (TASK_SIZE - len >= addr &&
172                     (!vma || addr + len <= vma->vm_start))
173                         return addr;
174         }
175
176         if (len > mm->cached_hole_size)
177                 start_addr = mm->free_area_cache;
178         else {
179                 start_addr = TASK_UNMAPPED_BASE;
180                 mm->cached_hole_size = 0;
181         }
182
183 full_search:
184         addr = ALIGN(start_addr, huge_page_size(h));
185
186         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
187                 /* At this point:  (!vma || addr < vma->vm_end). */
188                 if (TASK_SIZE - len < addr) {
189                         /*
190                          * Start a new search - just in case we missed
191                          * some holes.
192                          */
193                         if (start_addr != TASK_UNMAPPED_BASE) {
194                                 start_addr = TASK_UNMAPPED_BASE;
195                                 mm->cached_hole_size = 0;
196                                 goto full_search;
197                         }
198                         return -ENOMEM;
199                 }
200
201                 if (!vma || addr + len <= vma->vm_start) {
202                         mm->free_area_cache = addr + len;
203                         return addr;
204                 }
205                 if (addr + mm->cached_hole_size < vma->vm_start)
206                         mm->cached_hole_size = vma->vm_start - addr;
207                 addr = ALIGN(vma->vm_end, huge_page_size(h));
208         }
209 }
210 #endif
211
212 static int
213 hugetlbfs_read_actor(struct page *page, unsigned long offset,
214                         char __user *buf, unsigned long count,
215                         unsigned long size)
216 {
217         char *kaddr;
218         unsigned long left, copied = 0;
219         int i, chunksize;
220
221         if (size > count)
222                 size = count;
223
224         /* Find which 4k chunk and offset with in that chunk */
225         i = offset >> PAGE_CACHE_SHIFT;
226         offset = offset & ~PAGE_CACHE_MASK;
227
228         while (size) {
229                 chunksize = PAGE_CACHE_SIZE;
230                 if (offset)
231                         chunksize -= offset;
232                 if (chunksize > size)
233                         chunksize = size;
234                 kaddr = kmap(&page[i]);
235                 left = __copy_to_user(buf, kaddr + offset, chunksize);
236                 kunmap(&page[i]);
237                 if (left) {
238                         copied += (chunksize - left);
239                         break;
240                 }
241                 offset = 0;
242                 size -= chunksize;
243                 buf += chunksize;
244                 copied += chunksize;
245                 i++;
246         }
247         return copied ? copied : -EFAULT;
248 }
249
250 /*
251  * Support for read() - Find the page attached to f_mapping and copy out the
252  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
253  * since it has PAGE_CACHE_SIZE assumptions.
254  */
255 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
256                               size_t len, loff_t *ppos)
257 {
258         struct hstate *h = hstate_file(filp);
259         struct address_space *mapping = filp->f_mapping;
260         struct inode *inode = mapping->host;
261         unsigned long index = *ppos >> huge_page_shift(h);
262         unsigned long offset = *ppos & ~huge_page_mask(h);
263         unsigned long end_index;
264         loff_t isize;
265         ssize_t retval = 0;
266
267         /* validate length */
268         if (len == 0)
269                 goto out;
270
271         for (;;) {
272                 struct page *page;
273                 unsigned long nr, ret;
274                 int ra;
275
276                 /* nr is the maximum number of bytes to copy from this page */
277                 nr = huge_page_size(h);
278                 isize = i_size_read(inode);
279                 if (!isize)
280                         goto out;
281                 end_index = (isize - 1) >> huge_page_shift(h);
282                 if (index >= end_index) {
283                         if (index > end_index)
284                                 goto out;
285                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
286                         if (nr <= offset)
287                                 goto out;
288                 }
289                 nr = nr - offset;
290
291                 /* Find the page */
292                 page = find_lock_page(mapping, index);
293                 if (unlikely(page == NULL)) {
294                         /*
295                          * We have a HOLE, zero out the user-buffer for the
296                          * length of the hole or request.
297                          */
298                         ret = len < nr ? len : nr;
299                         if (clear_user(buf, ret))
300                                 ra = -EFAULT;
301                         else
302                                 ra = 0;
303                 } else {
304                         unlock_page(page);
305
306                         /*
307                          * We have the page, copy it to user space buffer.
308                          */
309                         ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
310                         ret = ra;
311                         page_cache_release(page);
312                 }
313                 if (ra < 0) {
314                         if (retval == 0)
315                                 retval = ra;
316                         goto out;
317                 }
318
319                 offset += ret;
320                 retval += ret;
321                 len -= ret;
322                 index += offset >> huge_page_shift(h);
323                 offset &= ~huge_page_mask(h);
324
325                 /* short read or no more work */
326                 if ((ret != nr) || (len == 0))
327                         break;
328         }
329 out:
330         *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
331         return retval;
332 }
333
334 static int hugetlbfs_write_begin(struct file *file,
335                         struct address_space *mapping,
336                         loff_t pos, unsigned len, unsigned flags,
337                         struct page **pagep, void **fsdata)
338 {
339         return -EINVAL;
340 }
341
342 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
343                         loff_t pos, unsigned len, unsigned copied,
344                         struct page *page, void *fsdata)
345 {
346         BUG();
347         return -EINVAL;
348 }
349
350 static void truncate_huge_page(struct page *page)
351 {
352         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
353         ClearPageUptodate(page);
354         delete_from_page_cache(page);
355 }
356
357 static void truncate_hugepages(struct inode *inode, loff_t lstart)
358 {
359         struct hstate *h = hstate_inode(inode);
360         struct address_space *mapping = &inode->i_data;
361         const pgoff_t start = lstart >> huge_page_shift(h);
362         struct pagevec pvec;
363         pgoff_t next;
364         int i, freed = 0;
365
366         pagevec_init(&pvec, 0);
367         next = start;
368         while (1) {
369                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
370                         if (next == start)
371                                 break;
372                         next = start;
373                         continue;
374                 }
375
376                 for (i = 0; i < pagevec_count(&pvec); ++i) {
377                         struct page *page = pvec.pages[i];
378
379                         lock_page(page);
380                         if (page->index > next)
381                                 next = page->index;
382                         ++next;
383                         truncate_huge_page(page);
384                         unlock_page(page);
385                         freed++;
386                 }
387                 huge_pagevec_release(&pvec);
388         }
389         BUG_ON(!lstart && mapping->nrpages);
390         hugetlb_unreserve_pages(inode, start, freed);
391 }
392
393 static void hugetlbfs_evict_inode(struct inode *inode)
394 {
395         truncate_hugepages(inode, 0);
396         clear_inode(inode);
397 }
398
399 static inline void
400 hugetlb_vmtruncate_list(struct rb_root *root, pgoff_t pgoff)
401 {
402         struct vm_area_struct *vma;
403
404         vma_interval_tree_foreach(vma, root, pgoff, ULONG_MAX) {
405                 unsigned long v_offset;
406
407                 /*
408                  * Can the expression below overflow on 32-bit arches?
409                  * No, because the interval tree returns us only those vmas
410                  * which overlap the truncated area starting at pgoff,
411                  * and no vma on a 32-bit arch can span beyond the 4GB.
412                  */
413                 if (vma->vm_pgoff < pgoff)
414                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
415                 else
416                         v_offset = 0;
417
418                 unmap_hugepage_range(vma, vma->vm_start + v_offset,
419                                      vma->vm_end, NULL);
420         }
421 }
422
423 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
424 {
425         pgoff_t pgoff;
426         struct address_space *mapping = inode->i_mapping;
427         struct hstate *h = hstate_inode(inode);
428
429         BUG_ON(offset & ~huge_page_mask(h));
430         pgoff = offset >> PAGE_SHIFT;
431
432         i_size_write(inode, offset);
433         mutex_lock(&mapping->i_mmap_mutex);
434         if (!RB_EMPTY_ROOT(&mapping->i_mmap))
435                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
436         mutex_unlock(&mapping->i_mmap_mutex);
437         truncate_hugepages(inode, offset);
438         return 0;
439 }
440
441 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
442 {
443         struct inode *inode = dentry->d_inode;
444         struct hstate *h = hstate_inode(inode);
445         int error;
446         unsigned int ia_valid = attr->ia_valid;
447
448         BUG_ON(!inode);
449
450         error = inode_change_ok(inode, attr);
451         if (error)
452                 return error;
453
454         if (ia_valid & ATTR_SIZE) {
455                 error = -EINVAL;
456                 if (attr->ia_size & ~huge_page_mask(h))
457                         return -EINVAL;
458                 error = hugetlb_vmtruncate(inode, attr->ia_size);
459                 if (error)
460                         return error;
461         }
462
463         setattr_copy(inode, attr);
464         mark_inode_dirty(inode);
465         return 0;
466 }
467
468 static struct inode *hugetlbfs_get_root(struct super_block *sb,
469                                         struct hugetlbfs_config *config)
470 {
471         struct inode *inode;
472
473         inode = new_inode(sb);
474         if (inode) {
475                 struct hugetlbfs_inode_info *info;
476                 inode->i_ino = get_next_ino();
477                 inode->i_mode = S_IFDIR | config->mode;
478                 inode->i_uid = config->uid;
479                 inode->i_gid = config->gid;
480                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
481                 info = HUGETLBFS_I(inode);
482                 mpol_shared_policy_init(&info->policy, NULL);
483                 inode->i_op = &hugetlbfs_dir_inode_operations;
484                 inode->i_fop = &simple_dir_operations;
485                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
486                 inc_nlink(inode);
487                 lockdep_annotate_inode_mutex_key(inode);
488         }
489         return inode;
490 }
491
492 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
493                                         struct inode *dir,
494                                         umode_t mode, dev_t dev)
495 {
496         struct inode *inode;
497
498         inode = new_inode(sb);
499         if (inode) {
500                 struct hugetlbfs_inode_info *info;
501                 inode->i_ino = get_next_ino();
502                 inode_init_owner(inode, dir, mode);
503                 inode->i_mapping->a_ops = &hugetlbfs_aops;
504                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
505                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
506                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
507                 info = HUGETLBFS_I(inode);
508                 /*
509                  * The policy is initialized here even if we are creating a
510                  * private inode because initialization simply creates an
511                  * an empty rb tree and calls spin_lock_init(), later when we
512                  * call mpol_free_shared_policy() it will just return because
513                  * the rb tree will still be empty.
514                  */
515                 mpol_shared_policy_init(&info->policy, NULL);
516                 switch (mode & S_IFMT) {
517                 default:
518                         init_special_inode(inode, mode, dev);
519                         break;
520                 case S_IFREG:
521                         inode->i_op = &hugetlbfs_inode_operations;
522                         inode->i_fop = &hugetlbfs_file_operations;
523                         break;
524                 case S_IFDIR:
525                         inode->i_op = &hugetlbfs_dir_inode_operations;
526                         inode->i_fop = &simple_dir_operations;
527
528                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
529                         inc_nlink(inode);
530                         break;
531                 case S_IFLNK:
532                         inode->i_op = &page_symlink_inode_operations;
533                         break;
534                 }
535                 lockdep_annotate_inode_mutex_key(inode);
536         }
537         return inode;
538 }
539
540 /*
541  * File creation. Allocate an inode, and we're done..
542  */
543 static int hugetlbfs_mknod(struct inode *dir,
544                         struct dentry *dentry, umode_t mode, dev_t dev)
545 {
546         struct inode *inode;
547         int error = -ENOSPC;
548
549         inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
550         if (inode) {
551                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
552                 d_instantiate(dentry, inode);
553                 dget(dentry);   /* Extra count - pin the dentry in core */
554                 error = 0;
555         }
556         return error;
557 }
558
559 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
560 {
561         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
562         if (!retval)
563                 inc_nlink(dir);
564         return retval;
565 }
566
567 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
568 {
569         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
570 }
571
572 static int hugetlbfs_symlink(struct inode *dir,
573                         struct dentry *dentry, const char *symname)
574 {
575         struct inode *inode;
576         int error = -ENOSPC;
577
578         inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
579         if (inode) {
580                 int l = strlen(symname)+1;
581                 error = page_symlink(inode, symname, l);
582                 if (!error) {
583                         d_instantiate(dentry, inode);
584                         dget(dentry);
585                 } else
586                         iput(inode);
587         }
588         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
589
590         return error;
591 }
592
593 /*
594  * mark the head page dirty
595  */
596 static int hugetlbfs_set_page_dirty(struct page *page)
597 {
598         struct page *head = compound_head(page);
599
600         SetPageDirty(head);
601         return 0;
602 }
603
604 static int hugetlbfs_migrate_page(struct address_space *mapping,
605                                 struct page *newpage, struct page *page,
606                                 enum migrate_mode mode)
607 {
608         int rc;
609
610         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
611         if (rc)
612                 return rc;
613         migrate_page_copy(newpage, page);
614
615         return 0;
616 }
617
618 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
619 {
620         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
621         struct hstate *h = hstate_inode(dentry->d_inode);
622
623         buf->f_type = HUGETLBFS_MAGIC;
624         buf->f_bsize = huge_page_size(h);
625         if (sbinfo) {
626                 spin_lock(&sbinfo->stat_lock);
627                 /* If no limits set, just report 0 for max/free/used
628                  * blocks, like simple_statfs() */
629                 if (sbinfo->spool) {
630                         long free_pages;
631
632                         spin_lock(&sbinfo->spool->lock);
633                         buf->f_blocks = sbinfo->spool->max_hpages;
634                         free_pages = sbinfo->spool->max_hpages
635                                 - sbinfo->spool->used_hpages;
636                         buf->f_bavail = buf->f_bfree = free_pages;
637                         spin_unlock(&sbinfo->spool->lock);
638                         buf->f_files = sbinfo->max_inodes;
639                         buf->f_ffree = sbinfo->free_inodes;
640                 }
641                 spin_unlock(&sbinfo->stat_lock);
642         }
643         buf->f_namelen = NAME_MAX;
644         return 0;
645 }
646
647 static void hugetlbfs_put_super(struct super_block *sb)
648 {
649         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
650
651         if (sbi) {
652                 sb->s_fs_info = NULL;
653
654                 if (sbi->spool)
655                         hugepage_put_subpool(sbi->spool);
656
657                 kfree(sbi);
658         }
659 }
660
661 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
662 {
663         if (sbinfo->free_inodes >= 0) {
664                 spin_lock(&sbinfo->stat_lock);
665                 if (unlikely(!sbinfo->free_inodes)) {
666                         spin_unlock(&sbinfo->stat_lock);
667                         return 0;
668                 }
669                 sbinfo->free_inodes--;
670                 spin_unlock(&sbinfo->stat_lock);
671         }
672
673         return 1;
674 }
675
676 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
677 {
678         if (sbinfo->free_inodes >= 0) {
679                 spin_lock(&sbinfo->stat_lock);
680                 sbinfo->free_inodes++;
681                 spin_unlock(&sbinfo->stat_lock);
682         }
683 }
684
685
686 static struct kmem_cache *hugetlbfs_inode_cachep;
687
688 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
689 {
690         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
691         struct hugetlbfs_inode_info *p;
692
693         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
694                 return NULL;
695         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
696         if (unlikely(!p)) {
697                 hugetlbfs_inc_free_inodes(sbinfo);
698                 return NULL;
699         }
700         return &p->vfs_inode;
701 }
702
703 static void hugetlbfs_i_callback(struct rcu_head *head)
704 {
705         struct inode *inode = container_of(head, struct inode, i_rcu);
706         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
707 }
708
709 static void hugetlbfs_destroy_inode(struct inode *inode)
710 {
711         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
712         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
713         call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
714 }
715
716 static const struct address_space_operations hugetlbfs_aops = {
717         .write_begin    = hugetlbfs_write_begin,
718         .write_end      = hugetlbfs_write_end,
719         .set_page_dirty = hugetlbfs_set_page_dirty,
720         .migratepage    = hugetlbfs_migrate_page,
721 };
722
723
724 static void init_once(void *foo)
725 {
726         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
727
728         inode_init_once(&ei->vfs_inode);
729 }
730
731 const struct file_operations hugetlbfs_file_operations = {
732         .read                   = hugetlbfs_read,
733         .mmap                   = hugetlbfs_file_mmap,
734         .fsync                  = noop_fsync,
735         .get_unmapped_area      = hugetlb_get_unmapped_area,
736         .llseek         = default_llseek,
737 };
738
739 static const struct inode_operations hugetlbfs_dir_inode_operations = {
740         .create         = hugetlbfs_create,
741         .lookup         = simple_lookup,
742         .link           = simple_link,
743         .unlink         = simple_unlink,
744         .symlink        = hugetlbfs_symlink,
745         .mkdir          = hugetlbfs_mkdir,
746         .rmdir          = simple_rmdir,
747         .mknod          = hugetlbfs_mknod,
748         .rename         = simple_rename,
749         .setattr        = hugetlbfs_setattr,
750 };
751
752 static const struct inode_operations hugetlbfs_inode_operations = {
753         .setattr        = hugetlbfs_setattr,
754 };
755
756 static const struct super_operations hugetlbfs_ops = {
757         .alloc_inode    = hugetlbfs_alloc_inode,
758         .destroy_inode  = hugetlbfs_destroy_inode,
759         .evict_inode    = hugetlbfs_evict_inode,
760         .statfs         = hugetlbfs_statfs,
761         .put_super      = hugetlbfs_put_super,
762         .show_options   = generic_show_options,
763 };
764
765 static int
766 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
767 {
768         char *p, *rest;
769         substring_t args[MAX_OPT_ARGS];
770         int option;
771         unsigned long long size = 0;
772         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
773
774         if (!options)
775                 return 0;
776
777         while ((p = strsep(&options, ",")) != NULL) {
778                 int token;
779                 if (!*p)
780                         continue;
781
782                 token = match_token(p, tokens, args);
783                 switch (token) {
784                 case Opt_uid:
785                         if (match_int(&args[0], &option))
786                                 goto bad_val;
787                         pconfig->uid = make_kuid(current_user_ns(), option);
788                         if (!uid_valid(pconfig->uid))
789                                 goto bad_val;
790                         break;
791
792                 case Opt_gid:
793                         if (match_int(&args[0], &option))
794                                 goto bad_val;
795                         pconfig->gid = make_kgid(current_user_ns(), option);
796                         if (!gid_valid(pconfig->gid))
797                                 goto bad_val;
798                         break;
799
800                 case Opt_mode:
801                         if (match_octal(&args[0], &option))
802                                 goto bad_val;
803                         pconfig->mode = option & 01777U;
804                         break;
805
806                 case Opt_size: {
807                         /* memparse() will accept a K/M/G without a digit */
808                         if (!isdigit(*args[0].from))
809                                 goto bad_val;
810                         size = memparse(args[0].from, &rest);
811                         setsize = SIZE_STD;
812                         if (*rest == '%')
813                                 setsize = SIZE_PERCENT;
814                         break;
815                 }
816
817                 case Opt_nr_inodes:
818                         /* memparse() will accept a K/M/G without a digit */
819                         if (!isdigit(*args[0].from))
820                                 goto bad_val;
821                         pconfig->nr_inodes = memparse(args[0].from, &rest);
822                         break;
823
824                 case Opt_pagesize: {
825                         unsigned long ps;
826                         ps = memparse(args[0].from, &rest);
827                         pconfig->hstate = size_to_hstate(ps);
828                         if (!pconfig->hstate) {
829                                 printk(KERN_ERR
830                                 "hugetlbfs: Unsupported page size %lu MB\n",
831                                         ps >> 20);
832                                 return -EINVAL;
833                         }
834                         break;
835                 }
836
837                 default:
838                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
839                                  p);
840                         return -EINVAL;
841                         break;
842                 }
843         }
844
845         /* Do size after hstate is set up */
846         if (setsize > NO_SIZE) {
847                 struct hstate *h = pconfig->hstate;
848                 if (setsize == SIZE_PERCENT) {
849                         size <<= huge_page_shift(h);
850                         size *= h->max_huge_pages;
851                         do_div(size, 100);
852                 }
853                 pconfig->nr_blocks = (size >> huge_page_shift(h));
854         }
855
856         return 0;
857
858 bad_val:
859         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
860                args[0].from, p);
861         return -EINVAL;
862 }
863
864 static int
865 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
866 {
867         int ret;
868         struct hugetlbfs_config config;
869         struct hugetlbfs_sb_info *sbinfo;
870
871         save_mount_options(sb, data);
872
873         config.nr_blocks = -1; /* No limit on size by default */
874         config.nr_inodes = -1; /* No limit on number of inodes by default */
875         config.uid = current_fsuid();
876         config.gid = current_fsgid();
877         config.mode = 0755;
878         config.hstate = &default_hstate;
879         ret = hugetlbfs_parse_options(data, &config);
880         if (ret)
881                 return ret;
882
883         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
884         if (!sbinfo)
885                 return -ENOMEM;
886         sb->s_fs_info = sbinfo;
887         sbinfo->hstate = config.hstate;
888         spin_lock_init(&sbinfo->stat_lock);
889         sbinfo->max_inodes = config.nr_inodes;
890         sbinfo->free_inodes = config.nr_inodes;
891         sbinfo->spool = NULL;
892         if (config.nr_blocks != -1) {
893                 sbinfo->spool = hugepage_new_subpool(config.nr_blocks);
894                 if (!sbinfo->spool)
895                         goto out_free;
896         }
897         sb->s_maxbytes = MAX_LFS_FILESIZE;
898         sb->s_blocksize = huge_page_size(config.hstate);
899         sb->s_blocksize_bits = huge_page_shift(config.hstate);
900         sb->s_magic = HUGETLBFS_MAGIC;
901         sb->s_op = &hugetlbfs_ops;
902         sb->s_time_gran = 1;
903         sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
904         if (!sb->s_root)
905                 goto out_free;
906         return 0;
907 out_free:
908         if (sbinfo->spool)
909                 kfree(sbinfo->spool);
910         kfree(sbinfo);
911         return -ENOMEM;
912 }
913
914 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
915         int flags, const char *dev_name, void *data)
916 {
917         return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
918 }
919
920 static struct file_system_type hugetlbfs_fs_type = {
921         .name           = "hugetlbfs",
922         .mount          = hugetlbfs_mount,
923         .kill_sb        = kill_litter_super,
924 };
925
926 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
927
928 static int can_do_hugetlb_shm(void)
929 {
930         kgid_t shm_group;
931         shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
932         return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
933 }
934
935 static int get_hstate_idx(int page_size_log)
936 {
937         struct hstate *h;
938
939         if (!page_size_log)
940                 return default_hstate_idx;
941         h = size_to_hstate(1 << page_size_log);
942         if (!h)
943                 return -1;
944         return h - hstates;
945 }
946
947 struct file *hugetlb_file_setup(const char *name, unsigned long addr,
948                                 size_t size, vm_flags_t acctflag,
949                                 struct user_struct **user,
950                                 int creat_flags, int page_size_log)
951 {
952         int error = -ENOMEM;
953         struct file *file;
954         struct inode *inode;
955         struct path path;
956         struct dentry *root;
957         struct qstr quick_string;
958         struct hstate *hstate;
959         unsigned long num_pages;
960         int hstate_idx;
961
962         hstate_idx = get_hstate_idx(page_size_log);
963         if (hstate_idx < 0)
964                 return ERR_PTR(-ENODEV);
965
966         *user = NULL;
967         if (!hugetlbfs_vfsmount[hstate_idx])
968                 return ERR_PTR(-ENOENT);
969
970         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
971                 *user = current_user();
972                 if (user_shm_lock(size, *user)) {
973                         task_lock(current);
974                         printk_once(KERN_WARNING
975                                 "%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
976                                 current->comm, current->pid);
977                         task_unlock(current);
978                 } else {
979                         *user = NULL;
980                         return ERR_PTR(-EPERM);
981                 }
982         }
983
984         root = hugetlbfs_vfsmount[hstate_idx]->mnt_root;
985         quick_string.name = name;
986         quick_string.len = strlen(quick_string.name);
987         quick_string.hash = 0;
988         path.dentry = d_alloc(root, &quick_string);
989         if (!path.dentry)
990                 goto out_shm_unlock;
991
992         path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]);
993         error = -ENOSPC;
994         inode = hugetlbfs_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0);
995         if (!inode)
996                 goto out_dentry;
997
998         hstate = hstate_inode(inode);
999         size += addr & ~huge_page_mask(hstate);
1000         num_pages = ALIGN(size, huge_page_size(hstate)) >>
1001                         huge_page_shift(hstate);
1002         error = -ENOMEM;
1003         if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag))
1004                 goto out_inode;
1005
1006         d_instantiate(path.dentry, inode);
1007         inode->i_size = size;
1008         clear_nlink(inode);
1009
1010         error = -ENFILE;
1011         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
1012                         &hugetlbfs_file_operations);
1013         if (!file)
1014                 goto out_dentry; /* inode is already attached */
1015
1016         return file;
1017
1018 out_inode:
1019         iput(inode);
1020 out_dentry:
1021         path_put(&path);
1022 out_shm_unlock:
1023         if (*user) {
1024                 user_shm_unlock(size, *user);
1025                 *user = NULL;
1026         }
1027         return ERR_PTR(error);
1028 }
1029
1030 static int __init init_hugetlbfs_fs(void)
1031 {
1032         struct hstate *h;
1033         int error;
1034         int i;
1035
1036         error = bdi_init(&hugetlbfs_backing_dev_info);
1037         if (error)
1038                 return error;
1039
1040         error = -ENOMEM;
1041         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1042                                         sizeof(struct hugetlbfs_inode_info),
1043                                         0, 0, init_once);
1044         if (hugetlbfs_inode_cachep == NULL)
1045                 goto out2;
1046
1047         error = register_filesystem(&hugetlbfs_fs_type);
1048         if (error)
1049                 goto out;
1050
1051         i = 0;
1052         for_each_hstate(h) {
1053                 char buf[50];
1054                 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
1055
1056                 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1057                 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1058                                                         buf);
1059
1060                 if (IS_ERR(hugetlbfs_vfsmount[i])) {
1061                         pr_err("hugetlb: Cannot mount internal hugetlbfs for "
1062                                 "page size %uK", ps_kb);
1063                         error = PTR_ERR(hugetlbfs_vfsmount[i]);
1064                         hugetlbfs_vfsmount[i] = NULL;
1065                 }
1066                 i++;
1067         }
1068         /* Non default hstates are optional */
1069         if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1070                 return 0;
1071
1072  out:
1073         kmem_cache_destroy(hugetlbfs_inode_cachep);
1074  out2:
1075         bdi_destroy(&hugetlbfs_backing_dev_info);
1076         return error;
1077 }
1078
1079 static void __exit exit_hugetlbfs_fs(void)
1080 {
1081         struct hstate *h;
1082         int i;
1083
1084
1085         /*
1086          * Make sure all delayed rcu free inodes are flushed before we
1087          * destroy cache.
1088          */
1089         rcu_barrier();
1090         kmem_cache_destroy(hugetlbfs_inode_cachep);
1091         i = 0;
1092         for_each_hstate(h)
1093                 kern_unmount(hugetlbfs_vfsmount[i++]);
1094         unregister_filesystem(&hugetlbfs_fs_type);
1095         bdi_destroy(&hugetlbfs_backing_dev_info);
1096 }
1097
1098 module_init(init_hugetlbfs_fs)
1099 module_exit(exit_hugetlbfs_fs)
1100
1101 MODULE_LICENSE("GPL");