2 * hugetlbpage-backed filesystem. Based on ramfs.
6 * Copyright (C) 2002 Linus Torvalds.
9 #include <linux/module.h>
10 #include <linux/thread_info.h>
11 #include <asm/current.h>
12 #include <linux/sched.h> /* remove ASAP */
14 #include <linux/mount.h>
15 #include <linux/file.h>
16 #include <linux/writeback.h>
17 #include <linux/pagemap.h>
18 #include <linux/highmem.h>
19 #include <linux/init.h>
20 #include <linux/string.h>
21 #include <linux/capability.h>
22 #include <linux/backing-dev.h>
23 #include <linux/hugetlb.h>
24 #include <linux/pagevec.h>
25 #include <linux/quotaops.h>
26 #include <linux/slab.h>
27 #include <linux/dnotify.h>
28 #include <linux/statfs.h>
29 #include <linux/security.h>
31 #include <asm/uaccess.h>
33 /* some random number */
34 #define HUGETLBFS_MAGIC 0x958458f6
36 static struct super_operations hugetlbfs_ops;
37 static const struct address_space_operations hugetlbfs_aops;
38 const struct file_operations hugetlbfs_file_operations;
39 static struct inode_operations hugetlbfs_dir_inode_operations;
40 static struct inode_operations hugetlbfs_inode_operations;
42 static struct backing_dev_info hugetlbfs_backing_dev_info = {
43 .ra_pages = 0, /* No readahead */
44 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
47 int sysctl_hugetlb_shm_group;
49 static void huge_pagevec_release(struct pagevec *pvec)
53 for (i = 0; i < pagevec_count(pvec); ++i)
54 put_page(pvec->pages[i]);
59 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
61 struct inode *inode = file->f_path.dentry->d_inode;
66 * vma alignment has already been checked by prepare_hugepage_range.
67 * If you add any error returns here, do so after setting VM_HUGETLB,
68 * so is_vm_hugetlb_page tests below unmap_region go the right way
69 * when do_mmap_pgoff unwinds (may be important on powerpc and ia64).
71 vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
72 vma->vm_ops = &hugetlb_vm_ops;
74 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
76 mutex_lock(&inode->i_mutex);
80 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
82 if (vma->vm_flags & VM_MAYSHARE &&
83 hugetlb_reserve_pages(inode, vma->vm_pgoff >> (HPAGE_SHIFT-PAGE_SHIFT),
88 hugetlb_prefault_arch_hook(vma->vm_mm);
89 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
92 mutex_unlock(&inode->i_mutex);
98 * Called under down_write(mmap_sem).
101 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
102 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
103 unsigned long len, unsigned long pgoff, unsigned long flags);
106 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
107 unsigned long len, unsigned long pgoff, unsigned long flags)
109 struct mm_struct *mm = current->mm;
110 struct vm_area_struct *vma;
111 unsigned long start_addr;
113 if (len & ~HPAGE_MASK)
119 addr = ALIGN(addr, HPAGE_SIZE);
120 vma = find_vma(mm, addr);
121 if (TASK_SIZE - len >= addr &&
122 (!vma || addr + len <= vma->vm_start))
126 start_addr = mm->free_area_cache;
128 if (len <= mm->cached_hole_size)
129 start_addr = TASK_UNMAPPED_BASE;
132 addr = ALIGN(start_addr, HPAGE_SIZE);
134 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
135 /* At this point: (!vma || addr < vma->vm_end). */
136 if (TASK_SIZE - len < addr) {
138 * Start a new search - just in case we missed
141 if (start_addr != TASK_UNMAPPED_BASE) {
142 start_addr = TASK_UNMAPPED_BASE;
148 if (!vma || addr + len <= vma->vm_start)
150 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
156 * Read a page. Again trivial. If it didn't already exist
157 * in the page cache, it is zero-filled.
159 static int hugetlbfs_readpage(struct file *file, struct page * page)
165 static int hugetlbfs_prepare_write(struct file *file,
166 struct page *page, unsigned offset, unsigned to)
171 static int hugetlbfs_commit_write(struct file *file,
172 struct page *page, unsigned offset, unsigned to)
177 static void truncate_huge_page(struct page *page)
179 cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
180 ClearPageUptodate(page);
181 remove_from_page_cache(page);
185 static void truncate_hugepages(struct inode *inode, loff_t lstart)
187 struct address_space *mapping = &inode->i_data;
188 const pgoff_t start = lstart >> HPAGE_SHIFT;
193 pagevec_init(&pvec, 0);
196 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
203 for (i = 0; i < pagevec_count(&pvec); ++i) {
204 struct page *page = pvec.pages[i];
207 if (page->index > next)
210 truncate_huge_page(page);
212 hugetlb_put_quota(mapping);
215 huge_pagevec_release(&pvec);
217 BUG_ON(!lstart && mapping->nrpages);
218 hugetlb_unreserve_pages(inode, start, freed);
221 static void hugetlbfs_delete_inode(struct inode *inode)
223 truncate_hugepages(inode, 0);
227 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
229 struct super_block *sb = inode->i_sb;
231 if (!hlist_unhashed(&inode->i_hash)) {
232 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
233 list_move(&inode->i_list, &inode_unused);
234 inodes_stat.nr_unused++;
235 if (!sb || (sb->s_flags & MS_ACTIVE)) {
236 spin_unlock(&inode_lock);
239 inode->i_state |= I_WILL_FREE;
240 spin_unlock(&inode_lock);
242 * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
243 * in our backing_dev_info.
245 write_inode_now(inode, 1);
246 spin_lock(&inode_lock);
247 inode->i_state &= ~I_WILL_FREE;
248 inodes_stat.nr_unused--;
249 hlist_del_init(&inode->i_hash);
251 list_del_init(&inode->i_list);
252 list_del_init(&inode->i_sb_list);
253 inode->i_state |= I_FREEING;
254 inodes_stat.nr_inodes--;
255 spin_unlock(&inode_lock);
256 truncate_hugepages(inode, 0);
258 destroy_inode(inode);
261 static void hugetlbfs_drop_inode(struct inode *inode)
264 generic_delete_inode(inode);
266 hugetlbfs_forget_inode(inode);
270 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
272 struct vm_area_struct *vma;
273 struct prio_tree_iter iter;
275 vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
276 unsigned long v_offset;
279 * Can the expression below overflow on 32-bit arches?
280 * No, because the prio_tree returns us only those vmas
281 * which overlap the truncated area starting at pgoff,
282 * and no vma on a 32-bit arch can span beyond the 4GB.
284 if (vma->vm_pgoff < pgoff)
285 v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
289 __unmap_hugepage_range(vma,
290 vma->vm_start + v_offset, vma->vm_end);
295 * Expanding truncates are not allowed.
297 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
300 struct address_space *mapping = inode->i_mapping;
302 if (offset > inode->i_size)
305 BUG_ON(offset & ~HPAGE_MASK);
306 pgoff = offset >> PAGE_SHIFT;
308 inode->i_size = offset;
309 spin_lock(&mapping->i_mmap_lock);
310 if (!prio_tree_empty(&mapping->i_mmap))
311 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
312 spin_unlock(&mapping->i_mmap_lock);
313 truncate_hugepages(inode, offset);
317 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
319 struct inode *inode = dentry->d_inode;
321 unsigned int ia_valid = attr->ia_valid;
325 error = inode_change_ok(inode, attr);
329 if (ia_valid & ATTR_SIZE) {
331 if (!(attr->ia_size & ~HPAGE_MASK))
332 error = hugetlb_vmtruncate(inode, attr->ia_size);
335 attr->ia_valid &= ~ATTR_SIZE;
337 error = inode_setattr(inode, attr);
342 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
343 gid_t gid, int mode, dev_t dev)
347 inode = new_inode(sb);
349 struct hugetlbfs_inode_info *info;
350 inode->i_mode = mode;
354 inode->i_mapping->a_ops = &hugetlbfs_aops;
355 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
356 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
357 INIT_LIST_HEAD(&inode->i_mapping->private_list);
358 info = HUGETLBFS_I(inode);
359 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
360 switch (mode & S_IFMT) {
362 init_special_inode(inode, mode, dev);
365 inode->i_op = &hugetlbfs_inode_operations;
366 inode->i_fop = &hugetlbfs_file_operations;
369 inode->i_op = &hugetlbfs_dir_inode_operations;
370 inode->i_fop = &simple_dir_operations;
372 /* directory inodes start off with i_nlink == 2 (for "." entry) */
376 inode->i_op = &page_symlink_inode_operations;
384 * File creation. Allocate an inode, and we're done..
386 static int hugetlbfs_mknod(struct inode *dir,
387 struct dentry *dentry, int mode, dev_t dev)
393 if (dir->i_mode & S_ISGID) {
398 gid = current->fsgid;
400 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
402 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
403 d_instantiate(dentry, inode);
404 dget(dentry); /* Extra count - pin the dentry in core */
410 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
412 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
418 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
420 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
423 static int hugetlbfs_symlink(struct inode *dir,
424 struct dentry *dentry, const char *symname)
430 if (dir->i_mode & S_ISGID)
433 gid = current->fsgid;
435 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
436 gid, S_IFLNK|S_IRWXUGO, 0);
438 int l = strlen(symname)+1;
439 error = page_symlink(inode, symname, l);
441 d_instantiate(dentry, inode);
446 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
452 * For direct-IO reads into hugetlb pages
454 static int hugetlbfs_set_page_dirty(struct page *page)
459 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
461 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
463 buf->f_type = HUGETLBFS_MAGIC;
464 buf->f_bsize = HPAGE_SIZE;
466 spin_lock(&sbinfo->stat_lock);
467 /* If no limits set, just report 0 for max/free/used
468 * blocks, like simple_statfs() */
469 if (sbinfo->max_blocks >= 0) {
470 buf->f_blocks = sbinfo->max_blocks;
471 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
472 buf->f_files = sbinfo->max_inodes;
473 buf->f_ffree = sbinfo->free_inodes;
475 spin_unlock(&sbinfo->stat_lock);
477 buf->f_namelen = NAME_MAX;
481 static void hugetlbfs_put_super(struct super_block *sb)
483 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
486 sb->s_fs_info = NULL;
491 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
493 if (sbinfo->free_inodes >= 0) {
494 spin_lock(&sbinfo->stat_lock);
495 if (unlikely(!sbinfo->free_inodes)) {
496 spin_unlock(&sbinfo->stat_lock);
499 sbinfo->free_inodes--;
500 spin_unlock(&sbinfo->stat_lock);
506 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
508 if (sbinfo->free_inodes >= 0) {
509 spin_lock(&sbinfo->stat_lock);
510 sbinfo->free_inodes++;
511 spin_unlock(&sbinfo->stat_lock);
516 static struct kmem_cache *hugetlbfs_inode_cachep;
518 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
520 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
521 struct hugetlbfs_inode_info *p;
523 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
525 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
527 hugetlbfs_inc_free_inodes(sbinfo);
530 return &p->vfs_inode;
533 static void hugetlbfs_destroy_inode(struct inode *inode)
535 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
536 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
537 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
540 static const struct address_space_operations hugetlbfs_aops = {
541 .readpage = hugetlbfs_readpage,
542 .prepare_write = hugetlbfs_prepare_write,
543 .commit_write = hugetlbfs_commit_write,
544 .set_page_dirty = hugetlbfs_set_page_dirty,
548 static void init_once(void *foo, struct kmem_cache *cachep, unsigned long flags)
550 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
552 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
553 SLAB_CTOR_CONSTRUCTOR)
554 inode_init_once(&ei->vfs_inode);
557 const struct file_operations hugetlbfs_file_operations = {
558 .mmap = hugetlbfs_file_mmap,
559 .fsync = simple_sync_file,
560 .get_unmapped_area = hugetlb_get_unmapped_area,
563 static struct inode_operations hugetlbfs_dir_inode_operations = {
564 .create = hugetlbfs_create,
565 .lookup = simple_lookup,
567 .unlink = simple_unlink,
568 .symlink = hugetlbfs_symlink,
569 .mkdir = hugetlbfs_mkdir,
570 .rmdir = simple_rmdir,
571 .mknod = hugetlbfs_mknod,
572 .rename = simple_rename,
573 .setattr = hugetlbfs_setattr,
576 static struct inode_operations hugetlbfs_inode_operations = {
577 .setattr = hugetlbfs_setattr,
580 static struct super_operations hugetlbfs_ops = {
581 .alloc_inode = hugetlbfs_alloc_inode,
582 .destroy_inode = hugetlbfs_destroy_inode,
583 .statfs = hugetlbfs_statfs,
584 .delete_inode = hugetlbfs_delete_inode,
585 .drop_inode = hugetlbfs_drop_inode,
586 .put_super = hugetlbfs_put_super,
590 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
592 char *opt, *value, *rest;
596 while ((opt = strsep(&options, ",")) != NULL) {
600 value = strchr(opt, '=');
601 if (!value || !*value)
606 if (!strcmp(opt, "uid"))
607 pconfig->uid = simple_strtoul(value, &value, 0);
608 else if (!strcmp(opt, "gid"))
609 pconfig->gid = simple_strtoul(value, &value, 0);
610 else if (!strcmp(opt, "mode"))
611 pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
612 else if (!strcmp(opt, "size")) {
613 unsigned long long size = memparse(value, &rest);
615 size <<= HPAGE_SHIFT;
616 size *= max_huge_pages;
620 pconfig->nr_blocks = (size >> HPAGE_SHIFT);
622 } else if (!strcmp(opt,"nr_inodes")) {
623 pconfig->nr_inodes = memparse(value, &rest);
635 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
637 struct inode * inode;
638 struct dentry * root;
640 struct hugetlbfs_config config;
641 struct hugetlbfs_sb_info *sbinfo;
643 config.nr_blocks = -1; /* No limit on size by default */
644 config.nr_inodes = -1; /* No limit on number of inodes by default */
645 config.uid = current->fsuid;
646 config.gid = current->fsgid;
648 ret = hugetlbfs_parse_options(data, &config);
653 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
656 sb->s_fs_info = sbinfo;
657 spin_lock_init(&sbinfo->stat_lock);
658 sbinfo->max_blocks = config.nr_blocks;
659 sbinfo->free_blocks = config.nr_blocks;
660 sbinfo->max_inodes = config.nr_inodes;
661 sbinfo->free_inodes = config.nr_inodes;
662 sb->s_maxbytes = MAX_LFS_FILESIZE;
663 sb->s_blocksize = HPAGE_SIZE;
664 sb->s_blocksize_bits = HPAGE_SHIFT;
665 sb->s_magic = HUGETLBFS_MAGIC;
666 sb->s_op = &hugetlbfs_ops;
668 inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
669 S_IFDIR | config.mode, 0);
673 root = d_alloc_root(inode);
685 int hugetlb_get_quota(struct address_space *mapping)
688 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
690 if (sbinfo->free_blocks > -1) {
691 spin_lock(&sbinfo->stat_lock);
692 if (sbinfo->free_blocks > 0)
693 sbinfo->free_blocks--;
696 spin_unlock(&sbinfo->stat_lock);
702 void hugetlb_put_quota(struct address_space *mapping)
704 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
706 if (sbinfo->free_blocks > -1) {
707 spin_lock(&sbinfo->stat_lock);
708 sbinfo->free_blocks++;
709 spin_unlock(&sbinfo->stat_lock);
713 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
714 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
716 return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
719 static struct file_system_type hugetlbfs_fs_type = {
721 .get_sb = hugetlbfs_get_sb,
722 .kill_sb = kill_litter_super,
725 static struct vfsmount *hugetlbfs_vfsmount;
727 static int can_do_hugetlb_shm(void)
729 return likely(capable(CAP_IPC_LOCK) ||
730 in_group_p(sysctl_hugetlb_shm_group) ||
734 struct file *hugetlb_zero_setup(size_t size)
739 struct dentry *dentry, *root;
740 struct qstr quick_string;
742 static atomic_t counter;
744 if (!can_do_hugetlb_shm())
745 return ERR_PTR(-EPERM);
747 if (!user_shm_lock(size, current->user))
748 return ERR_PTR(-ENOMEM);
750 root = hugetlbfs_vfsmount->mnt_root;
751 snprintf(buf, 16, "%u", atomic_inc_return(&counter));
752 quick_string.name = buf;
753 quick_string.len = strlen(quick_string.name);
754 quick_string.hash = 0;
755 dentry = d_alloc(root, &quick_string);
760 file = get_empty_filp();
765 inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
766 current->fsgid, S_IFREG | S_IRWXUGO, 0);
771 if (hugetlb_reserve_pages(inode, 0, size >> HPAGE_SHIFT))
774 d_instantiate(dentry, inode);
775 inode->i_size = size;
777 file->f_path.mnt = mntget(hugetlbfs_vfsmount);
778 file->f_path.dentry = dentry;
779 file->f_mapping = inode->i_mapping;
780 file->f_op = &hugetlbfs_file_operations;
781 file->f_mode = FMODE_WRITE | FMODE_READ;
791 user_shm_unlock(size, current->user);
792 return ERR_PTR(error);
795 static int __init init_hugetlbfs_fs(void)
798 struct vfsmount *vfsmount;
800 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
801 sizeof(struct hugetlbfs_inode_info),
802 0, 0, init_once, NULL);
803 if (hugetlbfs_inode_cachep == NULL)
806 error = register_filesystem(&hugetlbfs_fs_type);
810 vfsmount = kern_mount(&hugetlbfs_fs_type);
812 if (!IS_ERR(vfsmount)) {
813 hugetlbfs_vfsmount = vfsmount;
817 error = PTR_ERR(vfsmount);
821 kmem_cache_destroy(hugetlbfs_inode_cachep);
825 static void __exit exit_hugetlbfs_fs(void)
827 kmem_cache_destroy(hugetlbfs_inode_cachep);
828 unregister_filesystem(&hugetlbfs_fs_type);
831 module_init(init_hugetlbfs_fs)
832 module_exit(exit_hugetlbfs_fs)
834 MODULE_LICENSE("GPL");