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 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
46 int sysctl_hugetlb_shm_group;
48 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
50 struct inode *inode = file->f_dentry->d_inode;
51 struct address_space *mapping = inode->i_mapping;
55 if ((vma->vm_flags & (VM_MAYSHARE | VM_WRITE)) == VM_WRITE)
58 if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
61 if (vma->vm_start & ~HPAGE_MASK)
64 if (vma->vm_end & ~HPAGE_MASK)
67 if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
70 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
74 vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
75 vma->vm_ops = &hugetlb_vm_ops;
78 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
79 if (!(vma->vm_flags & VM_WRITE) && len > inode->i_size)
82 ret = hugetlb_prefault(mapping, vma);
86 if (inode->i_size < len)
95 * Called under down_write(mmap_sem).
98 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
99 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
100 unsigned long len, unsigned long pgoff, unsigned long flags);
103 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
104 unsigned long len, unsigned long pgoff, unsigned long flags)
106 struct mm_struct *mm = current->mm;
107 struct vm_area_struct *vma;
108 unsigned long start_addr;
110 if (len & ~HPAGE_MASK)
116 addr = ALIGN(addr, HPAGE_SIZE);
117 vma = find_vma(mm, addr);
118 if (TASK_SIZE - len >= addr &&
119 (!vma || addr + len <= vma->vm_start))
123 start_addr = mm->free_area_cache;
125 if (len <= mm->cached_hole_size)
126 start_addr = TASK_UNMAPPED_BASE;
129 addr = ALIGN(start_addr, HPAGE_SIZE);
131 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
132 /* At this point: (!vma || addr < vma->vm_end). */
133 if (TASK_SIZE - len < addr) {
135 * Start a new search - just in case we missed
138 if (start_addr != TASK_UNMAPPED_BASE) {
139 start_addr = TASK_UNMAPPED_BASE;
145 if (!vma || addr + len <= vma->vm_start)
147 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
153 * Read a page. Again trivial. If it didn't already exist
154 * in the page cache, it is zero-filled.
156 static int hugetlbfs_readpage(struct file *file, struct page * page)
162 static int hugetlbfs_prepare_write(struct file *file,
163 struct page *page, unsigned offset, unsigned to)
168 static int hugetlbfs_commit_write(struct file *file,
169 struct page *page, unsigned offset, unsigned to)
174 static void huge_pagevec_release(struct pagevec *pvec)
178 for (i = 0; i < pagevec_count(pvec); ++i)
179 put_page(pvec->pages[i]);
181 pagevec_reinit(pvec);
184 static void truncate_huge_page(struct page *page)
186 clear_page_dirty(page);
187 ClearPageUptodate(page);
188 remove_from_page_cache(page);
192 static void truncate_hugepages(struct address_space *mapping, loff_t lstart)
194 const pgoff_t start = lstart >> HPAGE_SHIFT;
199 pagevec_init(&pvec, 0);
202 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
209 for (i = 0; i < pagevec_count(&pvec); ++i) {
210 struct page *page = pvec.pages[i];
213 if (page->index > next)
216 truncate_huge_page(page);
218 hugetlb_put_quota(mapping);
220 huge_pagevec_release(&pvec);
222 BUG_ON(!lstart && mapping->nrpages);
225 static void hugetlbfs_delete_inode(struct inode *inode)
227 if (inode->i_data.nrpages)
228 truncate_hugepages(&inode->i_data, 0);
232 static void hugetlbfs_do_delete_inode(struct inode *inode)
234 struct super_operations *op = inode->i_sb->s_op;
236 list_del_init(&inode->i_list);
237 list_del_init(&inode->i_sb_list);
238 inode->i_state |= I_FREEING;
239 inodes_stat.nr_inodes--;
240 spin_unlock(&inode_lock);
242 security_inode_delete(inode);
244 if (op->delete_inode) {
245 void (*delete)(struct inode *) = op->delete_inode;
246 if (!is_bad_inode(inode))
248 /* Filesystems implementing their own
249 * s_op->delete_inode are required to call
250 * truncate_inode_pages and clear_inode()
255 truncate_inode_pages(&inode->i_data, 0);
259 spin_lock(&inode_lock);
260 hlist_del_init(&inode->i_hash);
261 spin_unlock(&inode_lock);
262 wake_up_inode(inode);
263 if (inode->i_state != I_CLEAR)
265 destroy_inode(inode);
268 static void hugetlbfs_forget_inode(struct inode *inode)
270 struct super_block *super_block = inode->i_sb;
272 if (hlist_unhashed(&inode->i_hash))
275 if (!(inode->i_state & (I_DIRTY|I_LOCK))) {
276 list_del(&inode->i_list);
277 list_add(&inode->i_list, &inode_unused);
279 inodes_stat.nr_unused++;
280 if (!super_block || (super_block->s_flags & MS_ACTIVE)) {
281 spin_unlock(&inode_lock);
285 /* write_inode_now() ? */
286 inodes_stat.nr_unused--;
287 hlist_del_init(&inode->i_hash);
289 list_del_init(&inode->i_list);
290 list_del_init(&inode->i_sb_list);
291 inode->i_state |= I_FREEING;
292 inodes_stat.nr_inodes--;
293 spin_unlock(&inode_lock);
294 if (inode->i_data.nrpages)
295 truncate_hugepages(&inode->i_data, 0);
298 destroy_inode(inode);
301 static void hugetlbfs_drop_inode(struct inode *inode)
304 hugetlbfs_do_delete_inode(inode);
306 hugetlbfs_forget_inode(inode);
310 * h_pgoff is in HPAGE_SIZE units.
311 * vma->vm_pgoff is in PAGE_SIZE units.
314 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
316 struct vm_area_struct *vma;
317 struct prio_tree_iter iter;
319 vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
320 unsigned long h_vm_pgoff;
321 unsigned long v_offset;
323 h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
324 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
326 * Is this VMA fully outside the truncation point?
328 if (h_vm_pgoff >= h_pgoff)
331 unmap_hugepage_range(vma,
332 vma->vm_start + v_offset, vma->vm_end);
337 * Expanding truncates are not allowed.
339 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
342 struct address_space *mapping = inode->i_mapping;
344 if (offset > inode->i_size)
347 BUG_ON(offset & ~HPAGE_MASK);
348 pgoff = offset >> HPAGE_SHIFT;
350 inode->i_size = offset;
351 spin_lock(&mapping->i_mmap_lock);
352 if (!prio_tree_empty(&mapping->i_mmap))
353 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
354 spin_unlock(&mapping->i_mmap_lock);
355 truncate_hugepages(mapping, offset);
359 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
361 struct inode *inode = dentry->d_inode;
363 unsigned int ia_valid = attr->ia_valid;
367 error = inode_change_ok(inode, attr);
371 if (ia_valid & ATTR_SIZE) {
373 if (!(attr->ia_size & ~HPAGE_MASK))
374 error = hugetlb_vmtruncate(inode, attr->ia_size);
377 attr->ia_valid &= ~ATTR_SIZE;
379 error = inode_setattr(inode, attr);
384 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
385 gid_t gid, int mode, dev_t dev)
389 inode = new_inode(sb);
391 struct hugetlbfs_inode_info *info;
392 inode->i_mode = mode;
395 inode->i_blksize = HPAGE_SIZE;
397 inode->i_mapping->a_ops = &hugetlbfs_aops;
398 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
399 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
400 info = HUGETLBFS_I(inode);
401 mpol_shared_policy_init(&info->policy);
402 switch (mode & S_IFMT) {
404 init_special_inode(inode, mode, dev);
407 inode->i_op = &hugetlbfs_inode_operations;
408 inode->i_fop = &hugetlbfs_file_operations;
411 inode->i_op = &hugetlbfs_dir_inode_operations;
412 inode->i_fop = &simple_dir_operations;
414 /* directory inodes start off with i_nlink == 2 (for "." entry) */
418 inode->i_op = &page_symlink_inode_operations;
426 * File creation. Allocate an inode, and we're done..
428 static int hugetlbfs_mknod(struct inode *dir,
429 struct dentry *dentry, int mode, dev_t dev)
435 if (dir->i_mode & S_ISGID) {
440 gid = current->fsgid;
442 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
444 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
445 d_instantiate(dentry, inode);
446 dget(dentry); /* Extra count - pin the dentry in core */
452 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
454 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
460 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
462 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
465 static int hugetlbfs_symlink(struct inode *dir,
466 struct dentry *dentry, const char *symname)
472 if (dir->i_mode & S_ISGID)
475 gid = current->fsgid;
477 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
478 gid, S_IFLNK|S_IRWXUGO, 0);
480 int l = strlen(symname)+1;
481 error = page_symlink(inode, symname, l);
483 d_instantiate(dentry, inode);
488 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
494 * For direct-IO reads into hugetlb pages
496 static int hugetlbfs_set_page_dirty(struct page *page)
501 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
503 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
505 buf->f_type = HUGETLBFS_MAGIC;
506 buf->f_bsize = HPAGE_SIZE;
508 spin_lock(&sbinfo->stat_lock);
509 buf->f_blocks = sbinfo->max_blocks;
510 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
511 buf->f_files = sbinfo->max_inodes;
512 buf->f_ffree = sbinfo->free_inodes;
513 spin_unlock(&sbinfo->stat_lock);
515 buf->f_namelen = NAME_MAX;
519 static void hugetlbfs_put_super(struct super_block *sb)
521 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
524 sb->s_fs_info = NULL;
529 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
531 if (sbinfo->free_inodes >= 0) {
532 spin_lock(&sbinfo->stat_lock);
533 if (unlikely(!sbinfo->free_inodes)) {
534 spin_unlock(&sbinfo->stat_lock);
537 sbinfo->free_inodes--;
538 spin_unlock(&sbinfo->stat_lock);
544 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
546 if (sbinfo->free_inodes >= 0) {
547 spin_lock(&sbinfo->stat_lock);
548 sbinfo->free_inodes++;
549 spin_unlock(&sbinfo->stat_lock);
554 static kmem_cache_t *hugetlbfs_inode_cachep;
556 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
558 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
559 struct hugetlbfs_inode_info *p;
561 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
563 p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
565 hugetlbfs_inc_free_inodes(sbinfo);
568 return &p->vfs_inode;
571 static void hugetlbfs_destroy_inode(struct inode *inode)
573 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
574 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
575 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
578 static struct address_space_operations hugetlbfs_aops = {
579 .readpage = hugetlbfs_readpage,
580 .prepare_write = hugetlbfs_prepare_write,
581 .commit_write = hugetlbfs_commit_write,
582 .set_page_dirty = hugetlbfs_set_page_dirty,
586 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
588 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
590 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
591 SLAB_CTOR_CONSTRUCTOR)
592 inode_init_once(&ei->vfs_inode);
595 struct file_operations hugetlbfs_file_operations = {
596 .mmap = hugetlbfs_file_mmap,
597 .fsync = simple_sync_file,
598 .get_unmapped_area = hugetlb_get_unmapped_area,
601 static struct inode_operations hugetlbfs_dir_inode_operations = {
602 .create = hugetlbfs_create,
603 .lookup = simple_lookup,
605 .unlink = simple_unlink,
606 .symlink = hugetlbfs_symlink,
607 .mkdir = hugetlbfs_mkdir,
608 .rmdir = simple_rmdir,
609 .mknod = hugetlbfs_mknod,
610 .rename = simple_rename,
611 .setattr = hugetlbfs_setattr,
614 static struct inode_operations hugetlbfs_inode_operations = {
615 .setattr = hugetlbfs_setattr,
618 static struct super_operations hugetlbfs_ops = {
619 .alloc_inode = hugetlbfs_alloc_inode,
620 .destroy_inode = hugetlbfs_destroy_inode,
621 .statfs = hugetlbfs_statfs,
622 .delete_inode = hugetlbfs_delete_inode,
623 .drop_inode = hugetlbfs_drop_inode,
624 .put_super = hugetlbfs_put_super,
628 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
630 char *opt, *value, *rest;
634 while ((opt = strsep(&options, ",")) != NULL) {
638 value = strchr(opt, '=');
639 if (!value || !*value)
644 if (!strcmp(opt, "uid"))
645 pconfig->uid = simple_strtoul(value, &value, 0);
646 else if (!strcmp(opt, "gid"))
647 pconfig->gid = simple_strtoul(value, &value, 0);
648 else if (!strcmp(opt, "mode"))
649 pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
650 else if (!strcmp(opt, "size")) {
651 unsigned long long size = memparse(value, &rest);
653 size <<= HPAGE_SHIFT;
654 size *= max_huge_pages;
659 pconfig->nr_blocks = (size >> HPAGE_SHIFT);
661 } else if (!strcmp(opt,"nr_inodes")) {
662 pconfig->nr_inodes = memparse(value, &rest);
674 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
676 struct inode * inode;
677 struct dentry * root;
679 struct hugetlbfs_config config;
680 struct hugetlbfs_sb_info *sbinfo;
682 config.nr_blocks = -1; /* No limit on size by default */
683 config.nr_inodes = -1; /* No limit on number of inodes by default */
684 config.uid = current->fsuid;
685 config.gid = current->fsgid;
687 ret = hugetlbfs_parse_options(data, &config);
692 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
695 sb->s_fs_info = sbinfo;
696 spin_lock_init(&sbinfo->stat_lock);
697 sbinfo->max_blocks = config.nr_blocks;
698 sbinfo->free_blocks = config.nr_blocks;
699 sbinfo->max_inodes = config.nr_inodes;
700 sbinfo->free_inodes = config.nr_inodes;
701 sb->s_maxbytes = MAX_LFS_FILESIZE;
702 sb->s_blocksize = HPAGE_SIZE;
703 sb->s_blocksize_bits = HPAGE_SHIFT;
704 sb->s_magic = HUGETLBFS_MAGIC;
705 sb->s_op = &hugetlbfs_ops;
707 inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
708 S_IFDIR | config.mode, 0);
712 root = d_alloc_root(inode);
724 int hugetlb_get_quota(struct address_space *mapping)
727 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
729 if (sbinfo->free_blocks > -1) {
730 spin_lock(&sbinfo->stat_lock);
731 if (sbinfo->free_blocks > 0)
732 sbinfo->free_blocks--;
735 spin_unlock(&sbinfo->stat_lock);
741 void hugetlb_put_quota(struct address_space *mapping)
743 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
745 if (sbinfo->free_blocks > -1) {
746 spin_lock(&sbinfo->stat_lock);
747 sbinfo->free_blocks++;
748 spin_unlock(&sbinfo->stat_lock);
752 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
753 int flags, const char *dev_name, void *data)
755 return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
758 static struct file_system_type hugetlbfs_fs_type = {
760 .get_sb = hugetlbfs_get_sb,
761 .kill_sb = kill_litter_super,
764 static struct vfsmount *hugetlbfs_vfsmount;
767 * Return the next identifier for a shm file
769 static unsigned long hugetlbfs_counter(void)
771 static DEFINE_SPINLOCK(lock);
772 static unsigned long counter;
781 static int can_do_hugetlb_shm(void)
783 return likely(capable(CAP_IPC_LOCK) ||
784 in_group_p(sysctl_hugetlb_shm_group) ||
788 struct file *hugetlb_zero_setup(size_t size)
793 struct dentry *dentry, *root;
794 struct qstr quick_string;
797 if (!can_do_hugetlb_shm())
798 return ERR_PTR(-EPERM);
800 if (!is_hugepage_mem_enough(size))
801 return ERR_PTR(-ENOMEM);
803 if (!user_shm_lock(size, current->user))
804 return ERR_PTR(-ENOMEM);
806 root = hugetlbfs_vfsmount->mnt_root;
807 snprintf(buf, 16, "%lu", hugetlbfs_counter());
808 quick_string.name = buf;
809 quick_string.len = strlen(quick_string.name);
810 quick_string.hash = 0;
811 dentry = d_alloc(root, &quick_string);
816 file = get_empty_filp();
821 inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
822 current->fsgid, S_IFREG | S_IRWXUGO, 0);
826 d_instantiate(dentry, inode);
827 inode->i_size = size;
829 file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
830 file->f_dentry = dentry;
831 file->f_mapping = inode->i_mapping;
832 file->f_op = &hugetlbfs_file_operations;
833 file->f_mode = FMODE_WRITE | FMODE_READ;
841 user_shm_unlock(size, current->user);
842 return ERR_PTR(error);
845 static int __init init_hugetlbfs_fs(void)
848 struct vfsmount *vfsmount;
850 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
851 sizeof(struct hugetlbfs_inode_info),
852 0, 0, init_once, NULL);
853 if (hugetlbfs_inode_cachep == NULL)
856 error = register_filesystem(&hugetlbfs_fs_type);
860 vfsmount = kern_mount(&hugetlbfs_fs_type);
862 if (!IS_ERR(vfsmount)) {
863 hugetlbfs_vfsmount = vfsmount;
867 error = PTR_ERR(vfsmount);
871 kmem_cache_destroy(hugetlbfs_inode_cachep);
875 static void __exit exit_hugetlbfs_fs(void)
877 kmem_cache_destroy(hugetlbfs_inode_cachep);
878 unregister_filesystem(&hugetlbfs_fs_type);
881 module_init(init_hugetlbfs_fs)
882 module_exit(exit_hugetlbfs_fs)
884 MODULE_LICENSE("GPL");