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/backing-dev.h>
22 #include <linux/hugetlb.h>
23 #include <linux/pagevec.h>
24 #include <linux/quotaops.h>
25 #include <linux/slab.h>
26 #include <linux/dnotify.h>
27 #include <linux/statfs.h>
28 #include <linux/security.h>
30 #include <asm/uaccess.h>
32 /* some random number */
33 #define HUGETLBFS_MAGIC 0x958458f6
35 static struct super_operations hugetlbfs_ops;
36 static struct address_space_operations hugetlbfs_aops;
37 struct file_operations hugetlbfs_file_operations;
38 static struct inode_operations hugetlbfs_dir_inode_operations;
39 static struct inode_operations hugetlbfs_inode_operations;
41 static struct backing_dev_info hugetlbfs_backing_dev_info = {
42 .ra_pages = 0, /* No readahead */
43 .memory_backed = 1, /* Does not contribute to dirty memory */
46 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
48 struct inode *inode = file->f_dentry->d_inode;
49 struct address_space *mapping = inode->i_mapping;
53 if (vma->vm_start & ~HPAGE_MASK)
56 if (vma->vm_end & ~HPAGE_MASK)
59 if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
62 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
66 vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
67 vma->vm_ops = &hugetlb_vm_ops;
68 ret = hugetlb_prefault(mapping, vma);
69 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
70 if (ret == 0 && inode->i_size < len)
78 * Called under down_write(mmap_sem), page_table_lock is not held
81 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
82 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
83 unsigned long len, unsigned long pgoff, unsigned long flags);
86 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
87 unsigned long len, unsigned long pgoff, unsigned long flags)
89 struct mm_struct *mm = current->mm;
90 struct vm_area_struct *vma;
92 if (len & ~HPAGE_MASK)
98 addr = ALIGN(addr, HPAGE_SIZE);
99 vma = find_vma(mm, addr);
100 if (TASK_SIZE - len >= addr &&
101 (!vma || addr + len <= vma->vm_start))
105 addr = ALIGN(mm->free_area_cache, HPAGE_SIZE);
107 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
108 /* At this point: (!vma || addr < vma->vm_end). */
109 if (TASK_SIZE - len < addr)
111 if (!vma || addr + len <= vma->vm_start)
113 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
119 * Read a page. Again trivial. If it didn't already exist
120 * in the page cache, it is zero-filled.
122 static int hugetlbfs_readpage(struct file *file, struct page * page)
128 static int hugetlbfs_prepare_write(struct file *file,
129 struct page *page, unsigned offset, unsigned to)
134 static int hugetlbfs_commit_write(struct file *file,
135 struct page *page, unsigned offset, unsigned to)
140 void huge_pagevec_release(struct pagevec *pvec)
144 for (i = 0; i < pagevec_count(pvec); ++i)
145 put_page(pvec->pages[i]);
147 pagevec_reinit(pvec);
150 void truncate_huge_page(struct page *page)
152 clear_page_dirty(page);
153 ClearPageUptodate(page);
154 remove_from_page_cache(page);
158 void truncate_hugepages(struct address_space *mapping, loff_t lstart)
160 const pgoff_t start = lstart >> HPAGE_SHIFT;
165 pagevec_init(&pvec, 0);
168 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
175 for (i = 0; i < pagevec_count(&pvec); ++i) {
176 struct page *page = pvec.pages[i];
179 if (page->index > next)
182 truncate_huge_page(page);
184 hugetlb_put_quota(mapping);
186 huge_pagevec_release(&pvec);
188 BUG_ON(!lstart && mapping->nrpages);
191 static void hugetlbfs_delete_inode(struct inode *inode)
193 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(inode->i_sb);
195 hlist_del_init(&inode->i_hash);
196 list_del_init(&inode->i_list);
197 inode->i_state |= I_FREEING;
198 inodes_stat.nr_inodes--;
199 spin_unlock(&inode_lock);
201 if (inode->i_data.nrpages)
202 truncate_hugepages(&inode->i_data, 0);
204 security_inode_delete(inode);
206 if (sbinfo->free_inodes >= 0) {
207 spin_lock(&sbinfo->stat_lock);
208 sbinfo->free_inodes++;
209 spin_unlock(&sbinfo->stat_lock);
213 destroy_inode(inode);
216 static void hugetlbfs_forget_inode(struct inode *inode)
218 struct super_block *super_block = inode->i_sb;
219 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(super_block);
221 if (hlist_unhashed(&inode->i_hash))
224 if (!(inode->i_state & (I_DIRTY|I_LOCK))) {
225 list_del(&inode->i_list);
226 list_add(&inode->i_list, &inode_unused);
228 inodes_stat.nr_unused++;
229 if (!super_block || (super_block->s_flags & MS_ACTIVE)) {
230 spin_unlock(&inode_lock);
234 /* write_inode_now() ? */
235 inodes_stat.nr_unused--;
236 hlist_del_init(&inode->i_hash);
238 list_del_init(&inode->i_list);
239 inode->i_state |= I_FREEING;
240 inodes_stat.nr_inodes--;
241 spin_unlock(&inode_lock);
242 if (inode->i_data.nrpages)
243 truncate_hugepages(&inode->i_data, 0);
245 if (sbinfo->free_inodes >= 0) {
246 spin_lock(&sbinfo->stat_lock);
247 sbinfo->free_inodes++;
248 spin_unlock(&sbinfo->stat_lock);
252 destroy_inode(inode);
255 static void hugetlbfs_drop_inode(struct inode *inode)
258 hugetlbfs_delete_inode(inode);
260 hugetlbfs_forget_inode(inode);
264 * h_pgoff is in HPAGE_SIZE units.
265 * vma->vm_pgoff is in PAGE_SIZE units.
268 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
270 struct vm_area_struct *vma = NULL;
271 struct prio_tree_iter iter;
273 while ((vma = vma_prio_tree_next(vma, root, &iter,
274 h_pgoff, ULONG_MAX)) != NULL) {
275 unsigned long h_vm_pgoff;
276 unsigned long v_length;
277 unsigned long v_offset;
279 h_vm_pgoff = vma->vm_pgoff << (HPAGE_SHIFT - PAGE_SHIFT);
280 v_length = vma->vm_end - vma->vm_start;
281 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
284 * Is this VMA fully outside the truncation point?
286 if (h_vm_pgoff >= h_pgoff)
289 zap_hugepage_range(vma,
290 vma->vm_start + v_offset,
291 v_length - v_offset);
296 * Expanding truncates are not allowed.
298 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
301 struct address_space *mapping = inode->i_mapping;
303 if (offset > inode->i_size)
306 BUG_ON(offset & ~HPAGE_MASK);
307 pgoff = offset >> HPAGE_SHIFT;
309 inode->i_size = offset;
310 spin_lock(&mapping->i_mmap_lock);
311 if (!prio_tree_empty(&mapping->i_mmap))
312 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
313 spin_unlock(&mapping->i_mmap_lock);
314 truncate_hugepages(mapping, offset);
318 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
320 struct inode *inode = dentry->d_inode;
322 unsigned int ia_valid = attr->ia_valid;
326 error = inode_change_ok(inode, attr);
330 error = security_inode_setattr(dentry, attr);
333 if (ia_valid & ATTR_SIZE) {
335 if (!(attr->ia_size & ~HPAGE_MASK))
336 error = hugetlb_vmtruncate(inode, attr->ia_size);
339 attr->ia_valid &= ~ATTR_SIZE;
341 error = inode_setattr(inode, attr);
346 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
347 gid_t gid, int mode, dev_t dev)
350 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
352 if (sbinfo->free_inodes >= 0) {
353 spin_lock(&sbinfo->stat_lock);
354 if (!sbinfo->free_inodes) {
355 spin_unlock(&sbinfo->stat_lock);
358 sbinfo->free_inodes--;
359 spin_unlock(&sbinfo->stat_lock);
362 inode = new_inode(sb);
364 struct hugetlbfs_inode_info *info;
365 inode->i_mode = mode;
368 inode->i_blksize = HPAGE_SIZE;
370 inode->i_mapping->a_ops = &hugetlbfs_aops;
371 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
372 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
373 info = HUGETLBFS_I(inode);
374 mpol_shared_policy_init(&info->policy);
375 switch (mode & S_IFMT) {
377 init_special_inode(inode, mode, dev);
380 inode->i_op = &hugetlbfs_inode_operations;
381 inode->i_fop = &hugetlbfs_file_operations;
384 inode->i_op = &hugetlbfs_dir_inode_operations;
385 inode->i_fop = &simple_dir_operations;
387 /* directory inodes start off with i_nlink == 2 (for "." entry) */
391 inode->i_op = &page_symlink_inode_operations;
399 * File creation. Allocate an inode, and we're done..
401 static int hugetlbfs_mknod(struct inode *dir,
402 struct dentry *dentry, int mode, dev_t dev)
408 if (dir->i_mode & S_ISGID) {
413 gid = current->fsgid;
415 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
417 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
418 d_instantiate(dentry, inode);
419 dget(dentry); /* Extra count - pin the dentry in core */
425 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
427 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
433 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
435 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
438 static int hugetlbfs_symlink(struct inode *dir,
439 struct dentry *dentry, const char *symname)
445 if (dir->i_mode & S_ISGID)
448 gid = current->fsgid;
450 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
451 gid, S_IFLNK|S_IRWXUGO, 0);
453 int l = strlen(symname)+1;
454 error = page_symlink(inode, symname, l);
456 d_instantiate(dentry, inode);
461 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
467 * For direct-IO reads into hugetlb pages
469 int hugetlbfs_set_page_dirty(struct page *page)
474 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
476 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
478 buf->f_type = HUGETLBFS_MAGIC;
479 buf->f_bsize = HPAGE_SIZE;
481 spin_lock(&sbinfo->stat_lock);
482 buf->f_blocks = sbinfo->max_blocks;
483 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
484 buf->f_files = sbinfo->max_inodes;
485 buf->f_ffree = sbinfo->free_inodes;
486 spin_unlock(&sbinfo->stat_lock);
488 buf->f_namelen = NAME_MAX;
492 static void hugetlbfs_put_super(struct super_block *sb)
494 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
497 sb->s_fs_info = NULL;
502 static kmem_cache_t *hugetlbfs_inode_cachep;
504 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
506 struct hugetlbfs_inode_info *p;
508 p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
511 return &p->vfs_inode;
514 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
516 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
518 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
519 SLAB_CTOR_CONSTRUCTOR)
520 inode_init_once(&ei->vfs_inode);
523 static void hugetlbfs_destroy_inode(struct inode *inode)
525 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
526 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
529 static struct address_space_operations hugetlbfs_aops = {
530 .readpage = hugetlbfs_readpage,
531 .prepare_write = hugetlbfs_prepare_write,
532 .commit_write = hugetlbfs_commit_write,
533 .set_page_dirty = hugetlbfs_set_page_dirty,
536 struct file_operations hugetlbfs_file_operations = {
537 .mmap = hugetlbfs_file_mmap,
538 .fsync = simple_sync_file,
539 .get_unmapped_area = hugetlb_get_unmapped_area,
542 static struct inode_operations hugetlbfs_dir_inode_operations = {
543 .create = hugetlbfs_create,
544 .lookup = simple_lookup,
546 .unlink = simple_unlink,
547 .symlink = hugetlbfs_symlink,
548 .mkdir = hugetlbfs_mkdir,
549 .rmdir = simple_rmdir,
550 .mknod = hugetlbfs_mknod,
551 .rename = simple_rename,
552 .setattr = hugetlbfs_setattr,
555 static struct inode_operations hugetlbfs_inode_operations = {
556 .setattr = hugetlbfs_setattr,
559 static struct super_operations hugetlbfs_ops = {
560 .alloc_inode = hugetlbfs_alloc_inode,
561 .destroy_inode = hugetlbfs_destroy_inode,
562 .statfs = hugetlbfs_statfs,
563 .drop_inode = hugetlbfs_drop_inode,
564 .put_super = hugetlbfs_put_super,
568 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
570 char *opt, *value, *rest;
574 while ((opt = strsep(&options, ",")) != NULL) {
578 value = strchr(opt, '=');
579 if (!value || !*value)
584 if (!strcmp(opt, "uid"))
585 pconfig->uid = simple_strtoul(value, &value, 0);
586 else if (!strcmp(opt, "gid"))
587 pconfig->gid = simple_strtoul(value, &value, 0);
588 else if (!strcmp(opt, "mode"))
589 pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
590 else if (!strcmp(opt, "size")) {
591 unsigned long long size = memparse(value, &rest);
593 size <<= HPAGE_SHIFT;
594 size *= max_huge_pages;
599 pconfig->nr_blocks = (size >> HPAGE_SHIFT);
601 } else if (!strcmp(opt,"nr_inodes")) {
602 pconfig->nr_inodes = memparse(value, &rest);
614 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
616 struct inode * inode;
617 struct dentry * root;
619 struct hugetlbfs_config config;
620 struct hugetlbfs_sb_info *sbinfo;
622 config.nr_blocks = -1; /* No limit on size by default */
623 config.nr_inodes = -1; /* No limit on number of inodes by default */
624 config.uid = current->fsuid;
625 config.gid = current->fsgid;
627 ret = hugetlbfs_parse_options(data, &config);
632 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
635 sb->s_fs_info = sbinfo;
636 spin_lock_init(&sbinfo->stat_lock);
637 sbinfo->max_blocks = config.nr_blocks;
638 sbinfo->free_blocks = config.nr_blocks;
639 sbinfo->max_inodes = config.nr_inodes;
640 sbinfo->free_inodes = config.nr_inodes;
641 sb->s_blocksize = HPAGE_SIZE;
642 sb->s_blocksize_bits = HPAGE_SHIFT;
643 sb->s_magic = HUGETLBFS_MAGIC;
644 sb->s_op = &hugetlbfs_ops;
645 inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
646 S_IFDIR | config.mode, 0);
650 root = d_alloc_root(inode);
662 int hugetlb_get_quota(struct address_space *mapping)
665 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
667 if (sbinfo->free_blocks > -1) {
668 spin_lock(&sbinfo->stat_lock);
669 if (sbinfo->free_blocks > 0)
670 sbinfo->free_blocks--;
673 spin_unlock(&sbinfo->stat_lock);
679 void hugetlb_put_quota(struct address_space *mapping)
681 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
683 if (sbinfo->free_blocks > -1) {
684 spin_lock(&sbinfo->stat_lock);
685 sbinfo->free_blocks++;
686 spin_unlock(&sbinfo->stat_lock);
690 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
691 int flags, const char *dev_name, void *data)
693 return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
696 static struct file_system_type hugetlbfs_fs_type = {
698 .get_sb = hugetlbfs_get_sb,
699 .kill_sb = kill_litter_super,
702 static struct vfsmount *hugetlbfs_vfsmount;
705 * Return the next identifier for a shm file
707 static unsigned long hugetlbfs_counter(void)
709 static spinlock_t lock = SPIN_LOCK_UNLOCKED;
710 static unsigned long counter;
719 struct file *hugetlb_zero_setup(size_t size)
724 struct dentry *dentry, *root;
725 struct qstr quick_string;
728 if (!capable(CAP_IPC_LOCK))
729 return ERR_PTR(-EPERM);
731 if (!is_hugepage_mem_enough(size))
732 return ERR_PTR(-ENOMEM);
734 root = hugetlbfs_vfsmount->mnt_root;
735 snprintf(buf, 16, "%lu", hugetlbfs_counter());
736 quick_string.name = buf;
737 quick_string.len = strlen(quick_string.name);
738 quick_string.hash = 0;
739 dentry = d_alloc(root, &quick_string);
741 return ERR_PTR(-ENOMEM);
744 file = get_empty_filp();
749 inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
750 current->fsgid, S_IFREG | S_IRWXUGO, 0);
754 d_instantiate(dentry, inode);
755 inode->i_size = size;
757 file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
758 file->f_dentry = dentry;
759 file->f_mapping = inode->i_mapping;
760 file->f_op = &hugetlbfs_file_operations;
761 file->f_mode = FMODE_WRITE | FMODE_READ;
768 return ERR_PTR(error);
771 static int __init init_hugetlbfs_fs(void)
774 struct vfsmount *vfsmount;
776 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
777 sizeof(struct hugetlbfs_inode_info),
778 0, SLAB_RECLAIM_ACCOUNT,
780 if (hugetlbfs_inode_cachep == NULL)
783 error = register_filesystem(&hugetlbfs_fs_type);
787 vfsmount = kern_mount(&hugetlbfs_fs_type);
789 if (!IS_ERR(vfsmount)) {
790 hugetlbfs_vfsmount = vfsmount;
794 error = PTR_ERR(vfsmount);
798 kmem_cache_destroy(hugetlbfs_inode_cachep);
802 static void __exit exit_hugetlbfs_fs(void)
804 kmem_cache_destroy(hugetlbfs_inode_cachep);
805 unregister_filesystem(&hugetlbfs_fs_type);
808 module_init(init_hugetlbfs_fs)
809 module_exit(exit_hugetlbfs_fs)
811 MODULE_LICENSE("GPL");