2 * hugetlbpage-backed filesystem. Based on ramfs.
4 * Nadia Yvette Chambers, 2002
6 * Copyright (C) 2002 Linus Torvalds.
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/module.h>
12 #include <linux/thread_info.h>
13 #include <asm/current.h>
14 #include <linux/sched.h> /* remove ASAP */
16 #include <linux/mount.h>
17 #include <linux/file.h>
18 #include <linux/kernel.h>
19 #include <linux/writeback.h>
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/init.h>
23 #include <linux/string.h>
24 #include <linux/capability.h>
25 #include <linux/ctype.h>
26 #include <linux/backing-dev.h>
27 #include <linux/hugetlb.h>
28 #include <linux/pagevec.h>
29 #include <linux/parser.h>
30 #include <linux/mman.h>
31 #include <linux/slab.h>
32 #include <linux/dnotify.h>
33 #include <linux/statfs.h>
34 #include <linux/security.h>
35 #include <linux/magic.h>
36 #include <linux/migrate.h>
38 #include <asm/uaccess.h>
40 static const struct super_operations hugetlbfs_ops;
41 static const struct address_space_operations hugetlbfs_aops;
42 const struct file_operations hugetlbfs_file_operations;
43 static const struct inode_operations hugetlbfs_dir_inode_operations;
44 static const struct inode_operations hugetlbfs_inode_operations;
46 struct hugetlbfs_config {
52 struct hstate *hstate;
56 struct hugetlbfs_inode_info {
57 struct shared_policy policy;
58 struct inode vfs_inode;
61 static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
63 return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
66 int sysctl_hugetlb_shm_group;
69 Opt_size, Opt_nr_inodes,
70 Opt_mode, Opt_uid, Opt_gid,
71 Opt_pagesize, Opt_min_size,
75 static const match_table_t tokens = {
76 {Opt_size, "size=%s"},
77 {Opt_nr_inodes, "nr_inodes=%s"},
78 {Opt_mode, "mode=%o"},
81 {Opt_pagesize, "pagesize=%s"},
82 {Opt_min_size, "min_size=%s"},
86 static void huge_pagevec_release(struct pagevec *pvec)
90 for (i = 0; i < pagevec_count(pvec); ++i)
91 put_page(pvec->pages[i]);
96 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
98 struct inode *inode = file_inode(file);
101 struct hstate *h = hstate_file(file);
104 * vma address alignment (but not the pgoff alignment) has
105 * already been checked by prepare_hugepage_range. If you add
106 * any error returns here, do so after setting VM_HUGETLB, so
107 * is_vm_hugetlb_page tests below unmap_region go the right
108 * way when do_mmap_pgoff unwinds (may be important on powerpc
111 vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
112 vma->vm_ops = &hugetlb_vm_ops;
114 if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
117 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
119 mutex_lock(&inode->i_mutex);
123 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
125 if (hugetlb_reserve_pages(inode,
126 vma->vm_pgoff >> huge_page_order(h),
127 len >> huge_page_shift(h), vma,
132 hugetlb_prefault_arch_hook(vma->vm_mm);
133 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
136 mutex_unlock(&inode->i_mutex);
142 * Called under down_write(mmap_sem).
145 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
147 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
148 unsigned long len, unsigned long pgoff, unsigned long flags)
150 struct mm_struct *mm = current->mm;
151 struct vm_area_struct *vma;
152 struct hstate *h = hstate_file(file);
153 struct vm_unmapped_area_info info;
155 if (len & ~huge_page_mask(h))
160 if (flags & MAP_FIXED) {
161 if (prepare_hugepage_range(file, addr, len))
167 addr = ALIGN(addr, huge_page_size(h));
168 vma = find_vma(mm, addr);
169 if (TASK_SIZE - len >= addr &&
170 (!vma || addr + len <= vma->vm_start))
176 info.low_limit = TASK_UNMAPPED_BASE;
177 info.high_limit = TASK_SIZE;
178 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
179 info.align_offset = 0;
180 return vm_unmapped_area(&info);
185 hugetlbfs_read_actor(struct page *page, unsigned long offset,
186 char __user *buf, unsigned long count,
190 unsigned long left, copied = 0;
196 /* Find which 4k chunk and offset with in that chunk */
197 i = offset >> PAGE_CACHE_SHIFT;
198 offset = offset & ~PAGE_CACHE_MASK;
201 chunksize = PAGE_CACHE_SIZE;
204 if (chunksize > size)
206 kaddr = kmap(&page[i]);
207 left = __copy_to_user(buf, kaddr + offset, chunksize);
210 copied += (chunksize - left);
219 return copied ? copied : -EFAULT;
223 * Support for read() - Find the page attached to f_mapping and copy out the
224 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
225 * since it has PAGE_CACHE_SIZE assumptions.
227 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
228 size_t len, loff_t *ppos)
230 struct hstate *h = hstate_file(filp);
231 struct address_space *mapping = filp->f_mapping;
232 struct inode *inode = mapping->host;
233 unsigned long index = *ppos >> huge_page_shift(h);
234 unsigned long offset = *ppos & ~huge_page_mask(h);
235 unsigned long end_index;
239 /* validate length */
245 unsigned long nr, ret;
248 /* nr is the maximum number of bytes to copy from this page */
249 nr = huge_page_size(h);
250 isize = i_size_read(inode);
253 end_index = (isize - 1) >> huge_page_shift(h);
254 if (index >= end_index) {
255 if (index > end_index)
257 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
264 page = find_lock_page(mapping, index);
265 if (unlikely(page == NULL)) {
267 * We have a HOLE, zero out the user-buffer for the
268 * length of the hole or request.
270 ret = len < nr ? len : nr;
271 if (clear_user(buf, ret))
279 * We have the page, copy it to user space buffer.
281 ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
283 page_cache_release(page);
294 index += offset >> huge_page_shift(h);
295 offset &= ~huge_page_mask(h);
297 /* short read or no more work */
298 if ((ret != nr) || (len == 0))
302 *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
306 static int hugetlbfs_write_begin(struct file *file,
307 struct address_space *mapping,
308 loff_t pos, unsigned len, unsigned flags,
309 struct page **pagep, void **fsdata)
314 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
315 loff_t pos, unsigned len, unsigned copied,
316 struct page *page, void *fsdata)
322 static void truncate_huge_page(struct page *page)
324 ClearPageDirty(page);
325 ClearPageUptodate(page);
326 delete_from_page_cache(page);
329 static void truncate_hugepages(struct inode *inode, loff_t lstart)
331 struct hstate *h = hstate_inode(inode);
332 struct address_space *mapping = &inode->i_data;
333 const pgoff_t start = lstart >> huge_page_shift(h);
338 pagevec_init(&pvec, 0);
341 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
348 for (i = 0; i < pagevec_count(&pvec); ++i) {
349 struct page *page = pvec.pages[i];
352 if (page->index > next)
355 truncate_huge_page(page);
359 huge_pagevec_release(&pvec);
361 BUG_ON(!lstart && mapping->nrpages);
362 hugetlb_unreserve_pages(inode, start, freed);
365 static void hugetlbfs_evict_inode(struct inode *inode)
367 struct resv_map *resv_map;
369 truncate_hugepages(inode, 0);
370 resv_map = (struct resv_map *)inode->i_mapping->private_data;
371 /* root inode doesn't have the resv_map, so we should check it */
373 resv_map_release(&resv_map->refs);
378 hugetlb_vmtruncate_list(struct rb_root *root, pgoff_t pgoff)
380 struct vm_area_struct *vma;
382 vma_interval_tree_foreach(vma, root, pgoff, ULONG_MAX) {
383 unsigned long v_offset;
386 * Can the expression below overflow on 32-bit arches?
387 * No, because the interval tree returns us only those vmas
388 * which overlap the truncated area starting at pgoff,
389 * and no vma on a 32-bit arch can span beyond the 4GB.
391 if (vma->vm_pgoff < pgoff)
392 v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
396 unmap_hugepage_range(vma, vma->vm_start + v_offset,
401 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
404 struct address_space *mapping = inode->i_mapping;
405 struct hstate *h = hstate_inode(inode);
407 BUG_ON(offset & ~huge_page_mask(h));
408 pgoff = offset >> PAGE_SHIFT;
410 i_size_write(inode, offset);
411 i_mmap_lock_write(mapping);
412 if (!RB_EMPTY_ROOT(&mapping->i_mmap))
413 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
414 i_mmap_unlock_write(mapping);
415 truncate_hugepages(inode, offset);
419 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
421 struct inode *inode = dentry->d_inode;
422 struct hstate *h = hstate_inode(inode);
424 unsigned int ia_valid = attr->ia_valid;
428 error = inode_change_ok(inode, attr);
432 if (ia_valid & ATTR_SIZE) {
434 if (attr->ia_size & ~huge_page_mask(h))
436 error = hugetlb_vmtruncate(inode, attr->ia_size);
441 setattr_copy(inode, attr);
442 mark_inode_dirty(inode);
446 static struct inode *hugetlbfs_get_root(struct super_block *sb,
447 struct hugetlbfs_config *config)
451 inode = new_inode(sb);
453 struct hugetlbfs_inode_info *info;
454 inode->i_ino = get_next_ino();
455 inode->i_mode = S_IFDIR | config->mode;
456 inode->i_uid = config->uid;
457 inode->i_gid = config->gid;
458 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
459 info = HUGETLBFS_I(inode);
460 mpol_shared_policy_init(&info->policy, NULL);
461 inode->i_op = &hugetlbfs_dir_inode_operations;
462 inode->i_fop = &simple_dir_operations;
463 /* directory inodes start off with i_nlink == 2 (for "." entry) */
465 lockdep_annotate_inode_mutex_key(inode);
471 * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
472 * be taken from reclaim -- unlike regular filesystems. This needs an
473 * annotation because huge_pmd_share() does an allocation under
476 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
478 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
480 umode_t mode, dev_t dev)
483 struct resv_map *resv_map;
485 resv_map = resv_map_alloc();
489 inode = new_inode(sb);
491 struct hugetlbfs_inode_info *info;
492 inode->i_ino = get_next_ino();
493 inode_init_owner(inode, dir, mode);
494 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
495 &hugetlbfs_i_mmap_rwsem_key);
496 inode->i_mapping->a_ops = &hugetlbfs_aops;
497 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
498 inode->i_mapping->private_data = resv_map;
499 info = HUGETLBFS_I(inode);
501 * The policy is initialized here even if we are creating a
502 * private inode because initialization simply creates an
503 * an empty rb tree and calls spin_lock_init(), later when we
504 * call mpol_free_shared_policy() it will just return because
505 * the rb tree will still be empty.
507 mpol_shared_policy_init(&info->policy, NULL);
508 switch (mode & S_IFMT) {
510 init_special_inode(inode, mode, dev);
513 inode->i_op = &hugetlbfs_inode_operations;
514 inode->i_fop = &hugetlbfs_file_operations;
517 inode->i_op = &hugetlbfs_dir_inode_operations;
518 inode->i_fop = &simple_dir_operations;
520 /* directory inodes start off with i_nlink == 2 (for "." entry) */
524 inode->i_op = &page_symlink_inode_operations;
527 lockdep_annotate_inode_mutex_key(inode);
529 kref_put(&resv_map->refs, resv_map_release);
535 * File creation. Allocate an inode, and we're done..
537 static int hugetlbfs_mknod(struct inode *dir,
538 struct dentry *dentry, umode_t mode, dev_t dev)
543 inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
545 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
546 d_instantiate(dentry, inode);
547 dget(dentry); /* Extra count - pin the dentry in core */
553 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
555 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
561 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
563 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
566 static int hugetlbfs_symlink(struct inode *dir,
567 struct dentry *dentry, const char *symname)
572 inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
574 int l = strlen(symname)+1;
575 error = page_symlink(inode, symname, l);
577 d_instantiate(dentry, inode);
582 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
588 * mark the head page dirty
590 static int hugetlbfs_set_page_dirty(struct page *page)
592 struct page *head = compound_head(page);
598 static int hugetlbfs_migrate_page(struct address_space *mapping,
599 struct page *newpage, struct page *page,
600 enum migrate_mode mode)
604 rc = migrate_huge_page_move_mapping(mapping, newpage, page);
605 if (rc != MIGRATEPAGE_SUCCESS)
607 migrate_page_copy(newpage, page);
609 return MIGRATEPAGE_SUCCESS;
612 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
614 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
615 struct hstate *h = hstate_inode(dentry->d_inode);
617 buf->f_type = HUGETLBFS_MAGIC;
618 buf->f_bsize = huge_page_size(h);
620 spin_lock(&sbinfo->stat_lock);
621 /* If no limits set, just report 0 for max/free/used
622 * blocks, like simple_statfs() */
626 spin_lock(&sbinfo->spool->lock);
627 buf->f_blocks = sbinfo->spool->max_hpages;
628 free_pages = sbinfo->spool->max_hpages
629 - sbinfo->spool->used_hpages;
630 buf->f_bavail = buf->f_bfree = free_pages;
631 spin_unlock(&sbinfo->spool->lock);
632 buf->f_files = sbinfo->max_inodes;
633 buf->f_ffree = sbinfo->free_inodes;
635 spin_unlock(&sbinfo->stat_lock);
637 buf->f_namelen = NAME_MAX;
641 static void hugetlbfs_put_super(struct super_block *sb)
643 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
646 sb->s_fs_info = NULL;
649 hugepage_put_subpool(sbi->spool);
655 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
657 if (sbinfo->free_inodes >= 0) {
658 spin_lock(&sbinfo->stat_lock);
659 if (unlikely(!sbinfo->free_inodes)) {
660 spin_unlock(&sbinfo->stat_lock);
663 sbinfo->free_inodes--;
664 spin_unlock(&sbinfo->stat_lock);
670 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
672 if (sbinfo->free_inodes >= 0) {
673 spin_lock(&sbinfo->stat_lock);
674 sbinfo->free_inodes++;
675 spin_unlock(&sbinfo->stat_lock);
680 static struct kmem_cache *hugetlbfs_inode_cachep;
682 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
684 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
685 struct hugetlbfs_inode_info *p;
687 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
689 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
691 hugetlbfs_inc_free_inodes(sbinfo);
694 return &p->vfs_inode;
697 static void hugetlbfs_i_callback(struct rcu_head *head)
699 struct inode *inode = container_of(head, struct inode, i_rcu);
700 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
703 static void hugetlbfs_destroy_inode(struct inode *inode)
705 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
706 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
707 call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
710 static const struct address_space_operations hugetlbfs_aops = {
711 .write_begin = hugetlbfs_write_begin,
712 .write_end = hugetlbfs_write_end,
713 .set_page_dirty = hugetlbfs_set_page_dirty,
714 .migratepage = hugetlbfs_migrate_page,
718 static void init_once(void *foo)
720 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
722 inode_init_once(&ei->vfs_inode);
725 const struct file_operations hugetlbfs_file_operations = {
726 .read = hugetlbfs_read,
727 .mmap = hugetlbfs_file_mmap,
729 .get_unmapped_area = hugetlb_get_unmapped_area,
730 .llseek = default_llseek,
733 static const struct inode_operations hugetlbfs_dir_inode_operations = {
734 .create = hugetlbfs_create,
735 .lookup = simple_lookup,
737 .unlink = simple_unlink,
738 .symlink = hugetlbfs_symlink,
739 .mkdir = hugetlbfs_mkdir,
740 .rmdir = simple_rmdir,
741 .mknod = hugetlbfs_mknod,
742 .rename = simple_rename,
743 .setattr = hugetlbfs_setattr,
746 static const struct inode_operations hugetlbfs_inode_operations = {
747 .setattr = hugetlbfs_setattr,
750 static const struct super_operations hugetlbfs_ops = {
751 .alloc_inode = hugetlbfs_alloc_inode,
752 .destroy_inode = hugetlbfs_destroy_inode,
753 .evict_inode = hugetlbfs_evict_inode,
754 .statfs = hugetlbfs_statfs,
755 .put_super = hugetlbfs_put_super,
756 .show_options = generic_show_options,
759 enum { NO_SIZE, SIZE_STD, SIZE_PERCENT };
762 * Convert size option passed from command line to number of huge pages
763 * in the pool specified by hstate. Size option could be in bytes
764 * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
767 hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
770 if (val_type == NO_SIZE)
773 if (val_type == SIZE_PERCENT) {
774 size_opt <<= huge_page_shift(h);
775 size_opt *= h->max_huge_pages;
776 do_div(size_opt, 100);
779 size_opt >>= huge_page_shift(h);
784 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
787 substring_t args[MAX_OPT_ARGS];
789 unsigned long long max_size_opt = 0, min_size_opt = 0;
790 int max_val_type = NO_SIZE, min_val_type = NO_SIZE;
795 while ((p = strsep(&options, ",")) != NULL) {
800 token = match_token(p, tokens, args);
803 if (match_int(&args[0], &option))
805 pconfig->uid = make_kuid(current_user_ns(), option);
806 if (!uid_valid(pconfig->uid))
811 if (match_int(&args[0], &option))
813 pconfig->gid = make_kgid(current_user_ns(), option);
814 if (!gid_valid(pconfig->gid))
819 if (match_octal(&args[0], &option))
821 pconfig->mode = option & 01777U;
825 /* memparse() will accept a K/M/G without a digit */
826 if (!isdigit(*args[0].from))
828 max_size_opt = memparse(args[0].from, &rest);
829 max_val_type = SIZE_STD;
831 max_val_type = SIZE_PERCENT;
836 /* memparse() will accept a K/M/G without a digit */
837 if (!isdigit(*args[0].from))
839 pconfig->nr_inodes = memparse(args[0].from, &rest);
844 ps = memparse(args[0].from, &rest);
845 pconfig->hstate = size_to_hstate(ps);
846 if (!pconfig->hstate) {
847 pr_err("Unsupported page size %lu MB\n",
855 /* memparse() will accept a K/M/G without a digit */
856 if (!isdigit(*args[0].from))
858 min_size_opt = memparse(args[0].from, &rest);
859 min_val_type = SIZE_STD;
861 min_val_type = SIZE_PERCENT;
866 pr_err("Bad mount option: \"%s\"\n", p);
873 * Use huge page pool size (in hstate) to convert the size
874 * options to number of huge pages. If NO_SIZE, -1 is returned.
876 pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
877 max_size_opt, max_val_type);
878 pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
879 min_size_opt, min_val_type);
882 * If max_size was specified, then min_size must be smaller
884 if (max_val_type > NO_SIZE &&
885 pconfig->min_hpages > pconfig->max_hpages) {
886 pr_err("minimum size can not be greater than maximum size\n");
893 pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p);
898 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
901 struct hugetlbfs_config config;
902 struct hugetlbfs_sb_info *sbinfo;
904 save_mount_options(sb, data);
906 config.max_hpages = -1; /* No limit on size by default */
907 config.nr_inodes = -1; /* No limit on number of inodes by default */
908 config.uid = current_fsuid();
909 config.gid = current_fsgid();
911 config.hstate = &default_hstate;
912 config.min_hpages = -1; /* No default minimum size */
913 ret = hugetlbfs_parse_options(data, &config);
917 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
920 sb->s_fs_info = sbinfo;
921 sbinfo->hstate = config.hstate;
922 spin_lock_init(&sbinfo->stat_lock);
923 sbinfo->max_inodes = config.nr_inodes;
924 sbinfo->free_inodes = config.nr_inodes;
925 sbinfo->spool = NULL;
927 * Allocate and initialize subpool if maximum or minimum size is
928 * specified. Any needed reservations (for minimim size) are taken
929 * taken when the subpool is created.
931 if (config.max_hpages != -1 || config.min_hpages != -1) {
932 sbinfo->spool = hugepage_new_subpool(config.hstate,
938 sb->s_maxbytes = MAX_LFS_FILESIZE;
939 sb->s_blocksize = huge_page_size(config.hstate);
940 sb->s_blocksize_bits = huge_page_shift(config.hstate);
941 sb->s_magic = HUGETLBFS_MAGIC;
942 sb->s_op = &hugetlbfs_ops;
944 sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
949 kfree(sbinfo->spool);
954 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
955 int flags, const char *dev_name, void *data)
957 return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
960 static struct file_system_type hugetlbfs_fs_type = {
962 .mount = hugetlbfs_mount,
963 .kill_sb = kill_litter_super,
965 MODULE_ALIAS_FS("hugetlbfs");
967 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
969 static int can_do_hugetlb_shm(void)
972 shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
973 return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
976 static int get_hstate_idx(int page_size_log)
978 struct hstate *h = hstate_sizelog(page_size_log);
985 static const struct dentry_operations anon_ops = {
986 .d_dname = simple_dname
990 * Note that size should be aligned to proper hugepage size in caller side,
991 * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
993 struct file *hugetlb_file_setup(const char *name, size_t size,
994 vm_flags_t acctflag, struct user_struct **user,
995 int creat_flags, int page_size_log)
997 struct file *file = ERR_PTR(-ENOMEM);
1000 struct super_block *sb;
1001 struct qstr quick_string;
1004 hstate_idx = get_hstate_idx(page_size_log);
1006 return ERR_PTR(-ENODEV);
1009 if (!hugetlbfs_vfsmount[hstate_idx])
1010 return ERR_PTR(-ENOENT);
1012 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
1013 *user = current_user();
1014 if (user_shm_lock(size, *user)) {
1016 pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
1017 current->comm, current->pid);
1018 task_unlock(current);
1021 return ERR_PTR(-EPERM);
1025 sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb;
1026 quick_string.name = name;
1027 quick_string.len = strlen(quick_string.name);
1028 quick_string.hash = 0;
1029 path.dentry = d_alloc_pseudo(sb, &quick_string);
1031 goto out_shm_unlock;
1033 d_set_d_op(path.dentry, &anon_ops);
1034 path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]);
1035 file = ERR_PTR(-ENOSPC);
1036 inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0);
1040 file = ERR_PTR(-ENOMEM);
1041 if (hugetlb_reserve_pages(inode, 0,
1042 size >> huge_page_shift(hstate_inode(inode)), NULL,
1046 d_instantiate(path.dentry, inode);
1047 inode->i_size = size;
1050 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
1051 &hugetlbfs_file_operations);
1053 goto out_dentry; /* inode is already attached */
1063 user_shm_unlock(size, *user);
1069 static int __init init_hugetlbfs_fs(void)
1075 if (!hugepages_supported()) {
1076 pr_info("disabling because there are no supported hugepage sizes\n");
1081 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1082 sizeof(struct hugetlbfs_inode_info),
1084 if (hugetlbfs_inode_cachep == NULL)
1087 error = register_filesystem(&hugetlbfs_fs_type);
1092 for_each_hstate(h) {
1094 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
1096 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1097 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1100 if (IS_ERR(hugetlbfs_vfsmount[i])) {
1101 pr_err("Cannot mount internal hugetlbfs for "
1102 "page size %uK", ps_kb);
1103 error = PTR_ERR(hugetlbfs_vfsmount[i]);
1104 hugetlbfs_vfsmount[i] = NULL;
1108 /* Non default hstates are optional */
1109 if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1113 kmem_cache_destroy(hugetlbfs_inode_cachep);
1118 static void __exit exit_hugetlbfs_fs(void)
1125 * Make sure all delayed rcu free inodes are flushed before we
1129 kmem_cache_destroy(hugetlbfs_inode_cachep);
1132 kern_unmount(hugetlbfs_vfsmount[i++]);
1133 unregister_filesystem(&hugetlbfs_fs_type);
1136 module_init(init_hugetlbfs_fs)
1137 module_exit(exit_hugetlbfs_fs)
1139 MODULE_LICENSE("GPL");