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_dentry->d_inode;
65 if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
68 if (vma->vm_start & ~HPAGE_MASK)
71 if (vma->vm_end & ~HPAGE_MASK)
74 if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
77 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
79 mutex_lock(&inode->i_mutex);
81 vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
82 vma->vm_ops = &hugetlb_vm_ops;
85 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
87 if (vma->vm_flags & VM_MAYSHARE &&
88 hugetlb_reserve_pages(inode, vma->vm_pgoff >> (HPAGE_SHIFT-PAGE_SHIFT),
93 hugetlb_prefault_arch_hook(vma->vm_mm);
94 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
97 mutex_unlock(&inode->i_mutex);
103 * Called under down_write(mmap_sem).
106 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
107 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
108 unsigned long len, unsigned long pgoff, unsigned long flags);
111 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
112 unsigned long len, unsigned long pgoff, unsigned long flags)
114 struct mm_struct *mm = current->mm;
115 struct vm_area_struct *vma;
116 unsigned long start_addr;
118 if (len & ~HPAGE_MASK)
124 addr = ALIGN(addr, HPAGE_SIZE);
125 vma = find_vma(mm, addr);
126 if (TASK_SIZE - len >= addr &&
127 (!vma || addr + len <= vma->vm_start))
131 start_addr = mm->free_area_cache;
133 if (len <= mm->cached_hole_size)
134 start_addr = TASK_UNMAPPED_BASE;
137 addr = ALIGN(start_addr, HPAGE_SIZE);
139 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
140 /* At this point: (!vma || addr < vma->vm_end). */
141 if (TASK_SIZE - len < addr) {
143 * Start a new search - just in case we missed
146 if (start_addr != TASK_UNMAPPED_BASE) {
147 start_addr = TASK_UNMAPPED_BASE;
153 if (!vma || addr + len <= vma->vm_start)
155 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
161 * Read a page. Again trivial. If it didn't already exist
162 * in the page cache, it is zero-filled.
164 static int hugetlbfs_readpage(struct file *file, struct page * page)
170 static int hugetlbfs_prepare_write(struct file *file,
171 struct page *page, unsigned offset, unsigned to)
176 static int hugetlbfs_commit_write(struct file *file,
177 struct page *page, unsigned offset, unsigned to)
182 static void truncate_huge_page(struct page *page)
184 clear_page_dirty(page);
185 ClearPageUptodate(page);
186 remove_from_page_cache(page);
190 static void truncate_hugepages(struct inode *inode, loff_t lstart)
192 struct address_space *mapping = &inode->i_data;
193 const pgoff_t start = lstart >> HPAGE_SHIFT;
198 pagevec_init(&pvec, 0);
201 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
208 for (i = 0; i < pagevec_count(&pvec); ++i) {
209 struct page *page = pvec.pages[i];
212 if (page->index > next)
215 truncate_huge_page(page);
217 hugetlb_put_quota(mapping);
220 huge_pagevec_release(&pvec);
222 BUG_ON(!lstart && mapping->nrpages);
223 hugetlb_unreserve_pages(inode, start, freed);
226 static void hugetlbfs_delete_inode(struct inode *inode)
228 truncate_hugepages(inode, 0);
232 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
234 struct super_block *sb = inode->i_sb;
236 if (!hlist_unhashed(&inode->i_hash)) {
237 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
238 list_move(&inode->i_list, &inode_unused);
239 inodes_stat.nr_unused++;
240 if (!sb || (sb->s_flags & MS_ACTIVE)) {
241 spin_unlock(&inode_lock);
244 inode->i_state |= I_WILL_FREE;
245 spin_unlock(&inode_lock);
247 * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
248 * in our backing_dev_info.
250 write_inode_now(inode, 1);
251 spin_lock(&inode_lock);
252 inode->i_state &= ~I_WILL_FREE;
253 inodes_stat.nr_unused--;
254 hlist_del_init(&inode->i_hash);
256 list_del_init(&inode->i_list);
257 list_del_init(&inode->i_sb_list);
258 inode->i_state |= I_FREEING;
259 inodes_stat.nr_inodes--;
260 spin_unlock(&inode_lock);
261 truncate_hugepages(inode, 0);
263 destroy_inode(inode);
266 static void hugetlbfs_drop_inode(struct inode *inode)
269 generic_delete_inode(inode);
271 hugetlbfs_forget_inode(inode);
275 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
277 struct vm_area_struct *vma;
278 struct prio_tree_iter iter;
280 vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
281 unsigned long v_offset;
284 * Can the expression below overflow on 32-bit arches?
285 * No, because the prio_tree returns us only those vmas
286 * which overlap the truncated area starting at pgoff,
287 * and no vma on a 32-bit arch can span beyond the 4GB.
289 if (vma->vm_pgoff < pgoff)
290 v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
294 __unmap_hugepage_range(vma,
295 vma->vm_start + v_offset, vma->vm_end);
300 * Expanding truncates are not allowed.
302 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
305 struct address_space *mapping = inode->i_mapping;
307 if (offset > inode->i_size)
310 BUG_ON(offset & ~HPAGE_MASK);
311 pgoff = offset >> PAGE_SHIFT;
313 inode->i_size = offset;
314 spin_lock(&mapping->i_mmap_lock);
315 if (!prio_tree_empty(&mapping->i_mmap))
316 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
317 spin_unlock(&mapping->i_mmap_lock);
318 truncate_hugepages(inode, offset);
322 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
324 struct inode *inode = dentry->d_inode;
326 unsigned int ia_valid = attr->ia_valid;
330 error = inode_change_ok(inode, attr);
334 if (ia_valid & ATTR_SIZE) {
336 if (!(attr->ia_size & ~HPAGE_MASK))
337 error = hugetlb_vmtruncate(inode, attr->ia_size);
340 attr->ia_valid &= ~ATTR_SIZE;
342 error = inode_setattr(inode, attr);
347 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
348 gid_t gid, int mode, dev_t dev)
352 inode = new_inode(sb);
354 struct hugetlbfs_inode_info *info;
355 inode->i_mode = mode;
359 inode->i_mapping->a_ops = &hugetlbfs_aops;
360 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
361 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
362 INIT_LIST_HEAD(&inode->i_mapping->private_list);
363 info = HUGETLBFS_I(inode);
364 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
365 switch (mode & S_IFMT) {
367 init_special_inode(inode, mode, dev);
370 inode->i_op = &hugetlbfs_inode_operations;
371 inode->i_fop = &hugetlbfs_file_operations;
374 inode->i_op = &hugetlbfs_dir_inode_operations;
375 inode->i_fop = &simple_dir_operations;
377 /* directory inodes start off with i_nlink == 2 (for "." entry) */
381 inode->i_op = &page_symlink_inode_operations;
389 * File creation. Allocate an inode, and we're done..
391 static int hugetlbfs_mknod(struct inode *dir,
392 struct dentry *dentry, int mode, dev_t dev)
398 if (dir->i_mode & S_ISGID) {
403 gid = current->fsgid;
405 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
407 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
408 d_instantiate(dentry, inode);
409 dget(dentry); /* Extra count - pin the dentry in core */
415 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
417 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
423 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
425 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
428 static int hugetlbfs_symlink(struct inode *dir,
429 struct dentry *dentry, const char *symname)
435 if (dir->i_mode & S_ISGID)
438 gid = current->fsgid;
440 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
441 gid, S_IFLNK|S_IRWXUGO, 0);
443 int l = strlen(symname)+1;
444 error = page_symlink(inode, symname, l);
446 d_instantiate(dentry, inode);
451 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
457 * For direct-IO reads into hugetlb pages
459 static int hugetlbfs_set_page_dirty(struct page *page)
464 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
466 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
468 buf->f_type = HUGETLBFS_MAGIC;
469 buf->f_bsize = HPAGE_SIZE;
471 spin_lock(&sbinfo->stat_lock);
472 /* If no limits set, just report 0 for max/free/used
473 * blocks, like simple_statfs() */
474 if (sbinfo->max_blocks >= 0) {
475 buf->f_blocks = sbinfo->max_blocks;
476 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
477 buf->f_files = sbinfo->max_inodes;
478 buf->f_ffree = sbinfo->free_inodes;
480 spin_unlock(&sbinfo->stat_lock);
482 buf->f_namelen = NAME_MAX;
486 static void hugetlbfs_put_super(struct super_block *sb)
488 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
491 sb->s_fs_info = NULL;
496 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
498 if (sbinfo->free_inodes >= 0) {
499 spin_lock(&sbinfo->stat_lock);
500 if (unlikely(!sbinfo->free_inodes)) {
501 spin_unlock(&sbinfo->stat_lock);
504 sbinfo->free_inodes--;
505 spin_unlock(&sbinfo->stat_lock);
511 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
513 if (sbinfo->free_inodes >= 0) {
514 spin_lock(&sbinfo->stat_lock);
515 sbinfo->free_inodes++;
516 spin_unlock(&sbinfo->stat_lock);
521 static kmem_cache_t *hugetlbfs_inode_cachep;
523 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
525 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
526 struct hugetlbfs_inode_info *p;
528 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
530 p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
532 hugetlbfs_inc_free_inodes(sbinfo);
535 return &p->vfs_inode;
538 static void hugetlbfs_destroy_inode(struct inode *inode)
540 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
541 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
542 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
545 static const struct address_space_operations hugetlbfs_aops = {
546 .readpage = hugetlbfs_readpage,
547 .prepare_write = hugetlbfs_prepare_write,
548 .commit_write = hugetlbfs_commit_write,
549 .set_page_dirty = hugetlbfs_set_page_dirty,
553 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
555 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
557 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
558 SLAB_CTOR_CONSTRUCTOR)
559 inode_init_once(&ei->vfs_inode);
562 const struct file_operations hugetlbfs_file_operations = {
563 .mmap = hugetlbfs_file_mmap,
564 .fsync = simple_sync_file,
565 .get_unmapped_area = hugetlb_get_unmapped_area,
568 static struct inode_operations hugetlbfs_dir_inode_operations = {
569 .create = hugetlbfs_create,
570 .lookup = simple_lookup,
572 .unlink = simple_unlink,
573 .symlink = hugetlbfs_symlink,
574 .mkdir = hugetlbfs_mkdir,
575 .rmdir = simple_rmdir,
576 .mknod = hugetlbfs_mknod,
577 .rename = simple_rename,
578 .setattr = hugetlbfs_setattr,
581 static struct inode_operations hugetlbfs_inode_operations = {
582 .setattr = hugetlbfs_setattr,
585 static struct super_operations hugetlbfs_ops = {
586 .alloc_inode = hugetlbfs_alloc_inode,
587 .destroy_inode = hugetlbfs_destroy_inode,
588 .statfs = hugetlbfs_statfs,
589 .delete_inode = hugetlbfs_delete_inode,
590 .drop_inode = hugetlbfs_drop_inode,
591 .put_super = hugetlbfs_put_super,
595 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
597 char *opt, *value, *rest;
601 while ((opt = strsep(&options, ",")) != NULL) {
605 value = strchr(opt, '=');
606 if (!value || !*value)
611 if (!strcmp(opt, "uid"))
612 pconfig->uid = simple_strtoul(value, &value, 0);
613 else if (!strcmp(opt, "gid"))
614 pconfig->gid = simple_strtoul(value, &value, 0);
615 else if (!strcmp(opt, "mode"))
616 pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
617 else if (!strcmp(opt, "size")) {
618 unsigned long long size = memparse(value, &rest);
620 size <<= HPAGE_SHIFT;
621 size *= max_huge_pages;
625 pconfig->nr_blocks = (size >> HPAGE_SHIFT);
627 } else if (!strcmp(opt,"nr_inodes")) {
628 pconfig->nr_inodes = memparse(value, &rest);
640 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
642 struct inode * inode;
643 struct dentry * root;
645 struct hugetlbfs_config config;
646 struct hugetlbfs_sb_info *sbinfo;
648 config.nr_blocks = -1; /* No limit on size by default */
649 config.nr_inodes = -1; /* No limit on number of inodes by default */
650 config.uid = current->fsuid;
651 config.gid = current->fsgid;
653 ret = hugetlbfs_parse_options(data, &config);
658 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
661 sb->s_fs_info = sbinfo;
662 spin_lock_init(&sbinfo->stat_lock);
663 sbinfo->max_blocks = config.nr_blocks;
664 sbinfo->free_blocks = config.nr_blocks;
665 sbinfo->max_inodes = config.nr_inodes;
666 sbinfo->free_inodes = config.nr_inodes;
667 sb->s_maxbytes = MAX_LFS_FILESIZE;
668 sb->s_blocksize = HPAGE_SIZE;
669 sb->s_blocksize_bits = HPAGE_SHIFT;
670 sb->s_magic = HUGETLBFS_MAGIC;
671 sb->s_op = &hugetlbfs_ops;
673 inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
674 S_IFDIR | config.mode, 0);
678 root = d_alloc_root(inode);
690 int hugetlb_get_quota(struct address_space *mapping)
693 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
695 if (sbinfo->free_blocks > -1) {
696 spin_lock(&sbinfo->stat_lock);
697 if (sbinfo->free_blocks > 0)
698 sbinfo->free_blocks--;
701 spin_unlock(&sbinfo->stat_lock);
707 void hugetlb_put_quota(struct address_space *mapping)
709 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
711 if (sbinfo->free_blocks > -1) {
712 spin_lock(&sbinfo->stat_lock);
713 sbinfo->free_blocks++;
714 spin_unlock(&sbinfo->stat_lock);
718 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
719 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
721 return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
724 static struct file_system_type hugetlbfs_fs_type = {
726 .get_sb = hugetlbfs_get_sb,
727 .kill_sb = kill_litter_super,
730 static struct vfsmount *hugetlbfs_vfsmount;
732 static int can_do_hugetlb_shm(void)
734 return likely(capable(CAP_IPC_LOCK) ||
735 in_group_p(sysctl_hugetlb_shm_group) ||
739 struct file *hugetlb_zero_setup(size_t size)
744 struct dentry *dentry, *root;
745 struct qstr quick_string;
747 static atomic_t counter;
749 if (!can_do_hugetlb_shm())
750 return ERR_PTR(-EPERM);
752 if (!user_shm_lock(size, current->user))
753 return ERR_PTR(-ENOMEM);
755 root = hugetlbfs_vfsmount->mnt_root;
756 snprintf(buf, 16, "%u", atomic_inc_return(&counter));
757 quick_string.name = buf;
758 quick_string.len = strlen(quick_string.name);
759 quick_string.hash = 0;
760 dentry = d_alloc(root, &quick_string);
765 file = get_empty_filp();
770 inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
771 current->fsgid, S_IFREG | S_IRWXUGO, 0);
776 if (hugetlb_reserve_pages(inode, 0, size >> HPAGE_SHIFT))
779 d_instantiate(dentry, inode);
780 inode->i_size = size;
782 file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
783 file->f_dentry = dentry;
784 file->f_mapping = inode->i_mapping;
785 file->f_op = &hugetlbfs_file_operations;
786 file->f_mode = FMODE_WRITE | FMODE_READ;
796 user_shm_unlock(size, current->user);
797 return ERR_PTR(error);
800 static int __init init_hugetlbfs_fs(void)
803 struct vfsmount *vfsmount;
805 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
806 sizeof(struct hugetlbfs_inode_info),
807 0, 0, init_once, NULL);
808 if (hugetlbfs_inode_cachep == NULL)
811 error = register_filesystem(&hugetlbfs_fs_type);
815 vfsmount = kern_mount(&hugetlbfs_fs_type);
817 if (!IS_ERR(vfsmount)) {
818 hugetlbfs_vfsmount = vfsmount;
822 error = PTR_ERR(vfsmount);
826 kmem_cache_destroy(hugetlbfs_inode_cachep);
830 static void __exit exit_hugetlbfs_fs(void)
832 kmem_cache_destroy(hugetlbfs_inode_cachep);
833 unregister_filesystem(&hugetlbfs_fs_type);
836 module_init(init_hugetlbfs_fs)
837 module_exit(exit_hugetlbfs_fs)
839 MODULE_LICENSE("GPL");