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/kernel.h>
17 #include <linux/writeback.h>
18 #include <linux/pagemap.h>
19 #include <linux/highmem.h>
20 #include <linux/init.h>
21 #include <linux/string.h>
22 #include <linux/capability.h>
23 #include <linux/ctype.h>
24 #include <linux/backing-dev.h>
25 #include <linux/hugetlb.h>
26 #include <linux/pagevec.h>
27 #include <linux/parser.h>
28 #include <linux/mman.h>
29 #include <linux/quotaops.h>
30 #include <linux/slab.h>
31 #include <linux/dnotify.h>
32 #include <linux/statfs.h>
33 #include <linux/security.h>
35 #include <asm/uaccess.h>
37 /* some random number */
38 #define HUGETLBFS_MAGIC 0x958458f6
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 static struct backing_dev_info hugetlbfs_backing_dev_info = {
47 .ra_pages = 0, /* No readahead */
48 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
51 int sysctl_hugetlb_shm_group;
54 Opt_size, Opt_nr_inodes,
55 Opt_mode, Opt_uid, Opt_gid,
59 static match_table_t tokens = {
60 {Opt_size, "size=%s"},
61 {Opt_nr_inodes, "nr_inodes=%s"},
62 {Opt_mode, "mode=%o"},
68 static void huge_pagevec_release(struct pagevec *pvec)
72 for (i = 0; i < pagevec_count(pvec); ++i)
73 put_page(pvec->pages[i]);
78 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
80 struct inode *inode = file->f_path.dentry->d_inode;
85 * vma address alignment (but not the pgoff alignment) has
86 * already been checked by prepare_hugepage_range. If you add
87 * any error returns here, do so after setting VM_HUGETLB, so
88 * is_vm_hugetlb_page tests below unmap_region go the right
89 * way when do_mmap_pgoff unwinds (may be important on powerpc
92 vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
93 vma->vm_ops = &hugetlb_vm_ops;
95 if (vma->vm_pgoff & ~(HPAGE_MASK >> PAGE_SHIFT))
98 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
100 mutex_lock(&inode->i_mutex);
104 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
106 if (vma->vm_flags & VM_MAYSHARE &&
107 hugetlb_reserve_pages(inode, vma->vm_pgoff >> (HPAGE_SHIFT-PAGE_SHIFT),
112 hugetlb_prefault_arch_hook(vma->vm_mm);
113 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
116 mutex_unlock(&inode->i_mutex);
122 * Called under down_write(mmap_sem).
125 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
127 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
128 unsigned long len, unsigned long pgoff, unsigned long flags)
130 struct mm_struct *mm = current->mm;
131 struct vm_area_struct *vma;
132 unsigned long start_addr;
134 if (len & ~HPAGE_MASK)
139 if (flags & MAP_FIXED) {
140 if (prepare_hugepage_range(addr, len))
146 addr = ALIGN(addr, HPAGE_SIZE);
147 vma = find_vma(mm, addr);
148 if (TASK_SIZE - len >= addr &&
149 (!vma || addr + len <= vma->vm_start))
153 start_addr = mm->free_area_cache;
155 if (len <= mm->cached_hole_size)
156 start_addr = TASK_UNMAPPED_BASE;
159 addr = ALIGN(start_addr, HPAGE_SIZE);
161 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
162 /* At this point: (!vma || addr < vma->vm_end). */
163 if (TASK_SIZE - len < addr) {
165 * Start a new search - just in case we missed
168 if (start_addr != TASK_UNMAPPED_BASE) {
169 start_addr = TASK_UNMAPPED_BASE;
175 if (!vma || addr + len <= vma->vm_start)
177 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
183 * Read a page. Again trivial. If it didn't already exist
184 * in the page cache, it is zero-filled.
186 static int hugetlbfs_readpage(struct file *file, struct page * page)
192 static int hugetlbfs_write_begin(struct file *file,
193 struct address_space *mapping,
194 loff_t pos, unsigned len, unsigned flags,
195 struct page **pagep, void **fsdata)
200 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
201 loff_t pos, unsigned len, unsigned copied,
202 struct page *page, void *fsdata)
208 static void truncate_huge_page(struct page *page)
210 cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
211 ClearPageUptodate(page);
212 remove_from_page_cache(page);
216 static void truncate_hugepages(struct inode *inode, loff_t lstart)
218 struct address_space *mapping = &inode->i_data;
219 const pgoff_t start = lstart >> HPAGE_SHIFT;
224 pagevec_init(&pvec, 0);
227 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
234 for (i = 0; i < pagevec_count(&pvec); ++i) {
235 struct page *page = pvec.pages[i];
238 if (page->index > next)
241 truncate_huge_page(page);
243 hugetlb_put_quota(mapping);
246 huge_pagevec_release(&pvec);
248 BUG_ON(!lstart && mapping->nrpages);
249 hugetlb_unreserve_pages(inode, start, freed);
252 static void hugetlbfs_delete_inode(struct inode *inode)
254 truncate_hugepages(inode, 0);
258 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
260 struct super_block *sb = inode->i_sb;
262 if (!hlist_unhashed(&inode->i_hash)) {
263 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
264 list_move(&inode->i_list, &inode_unused);
265 inodes_stat.nr_unused++;
266 if (!sb || (sb->s_flags & MS_ACTIVE)) {
267 spin_unlock(&inode_lock);
270 inode->i_state |= I_WILL_FREE;
271 spin_unlock(&inode_lock);
273 * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
274 * in our backing_dev_info.
276 write_inode_now(inode, 1);
277 spin_lock(&inode_lock);
278 inode->i_state &= ~I_WILL_FREE;
279 inodes_stat.nr_unused--;
280 hlist_del_init(&inode->i_hash);
282 list_del_init(&inode->i_list);
283 list_del_init(&inode->i_sb_list);
284 inode->i_state |= I_FREEING;
285 inodes_stat.nr_inodes--;
286 spin_unlock(&inode_lock);
287 truncate_hugepages(inode, 0);
289 destroy_inode(inode);
292 static void hugetlbfs_drop_inode(struct inode *inode)
295 generic_delete_inode(inode);
297 hugetlbfs_forget_inode(inode);
301 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
303 struct vm_area_struct *vma;
304 struct prio_tree_iter iter;
306 vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
307 unsigned long v_offset;
310 * Can the expression below overflow on 32-bit arches?
311 * No, because the prio_tree returns us only those vmas
312 * which overlap the truncated area starting at pgoff,
313 * and no vma on a 32-bit arch can span beyond the 4GB.
315 if (vma->vm_pgoff < pgoff)
316 v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
320 __unmap_hugepage_range(vma,
321 vma->vm_start + v_offset, vma->vm_end);
326 * Expanding truncates are not allowed.
328 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
331 struct address_space *mapping = inode->i_mapping;
333 if (offset > inode->i_size)
336 BUG_ON(offset & ~HPAGE_MASK);
337 pgoff = offset >> PAGE_SHIFT;
339 inode->i_size = offset;
340 spin_lock(&mapping->i_mmap_lock);
341 if (!prio_tree_empty(&mapping->i_mmap))
342 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
343 spin_unlock(&mapping->i_mmap_lock);
344 truncate_hugepages(inode, offset);
348 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
350 struct inode *inode = dentry->d_inode;
352 unsigned int ia_valid = attr->ia_valid;
356 error = inode_change_ok(inode, attr);
360 if (ia_valid & ATTR_SIZE) {
362 if (!(attr->ia_size & ~HPAGE_MASK))
363 error = hugetlb_vmtruncate(inode, attr->ia_size);
366 attr->ia_valid &= ~ATTR_SIZE;
368 error = inode_setattr(inode, attr);
373 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
374 gid_t gid, int mode, dev_t dev)
378 inode = new_inode(sb);
380 struct hugetlbfs_inode_info *info;
381 inode->i_mode = mode;
385 inode->i_mapping->a_ops = &hugetlbfs_aops;
386 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
387 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
388 INIT_LIST_HEAD(&inode->i_mapping->private_list);
389 info = HUGETLBFS_I(inode);
390 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
391 switch (mode & S_IFMT) {
393 init_special_inode(inode, mode, dev);
396 inode->i_op = &hugetlbfs_inode_operations;
397 inode->i_fop = &hugetlbfs_file_operations;
400 inode->i_op = &hugetlbfs_dir_inode_operations;
401 inode->i_fop = &simple_dir_operations;
403 /* directory inodes start off with i_nlink == 2 (for "." entry) */
407 inode->i_op = &page_symlink_inode_operations;
415 * File creation. Allocate an inode, and we're done..
417 static int hugetlbfs_mknod(struct inode *dir,
418 struct dentry *dentry, int mode, dev_t dev)
424 if (dir->i_mode & S_ISGID) {
429 gid = current->fsgid;
431 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
433 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
434 d_instantiate(dentry, inode);
435 dget(dentry); /* Extra count - pin the dentry in core */
441 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
443 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
449 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
451 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
454 static int hugetlbfs_symlink(struct inode *dir,
455 struct dentry *dentry, const char *symname)
461 if (dir->i_mode & S_ISGID)
464 gid = current->fsgid;
466 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
467 gid, S_IFLNK|S_IRWXUGO, 0);
469 int l = strlen(symname)+1;
470 error = page_symlink(inode, symname, l);
472 d_instantiate(dentry, inode);
477 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
483 * mark the head page dirty
485 static int hugetlbfs_set_page_dirty(struct page *page)
487 struct page *head = compound_head(page);
493 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
495 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
497 buf->f_type = HUGETLBFS_MAGIC;
498 buf->f_bsize = HPAGE_SIZE;
500 spin_lock(&sbinfo->stat_lock);
501 /* If no limits set, just report 0 for max/free/used
502 * blocks, like simple_statfs() */
503 if (sbinfo->max_blocks >= 0) {
504 buf->f_blocks = sbinfo->max_blocks;
505 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
506 buf->f_files = sbinfo->max_inodes;
507 buf->f_ffree = sbinfo->free_inodes;
509 spin_unlock(&sbinfo->stat_lock);
511 buf->f_namelen = NAME_MAX;
515 static void hugetlbfs_put_super(struct super_block *sb)
517 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
520 sb->s_fs_info = NULL;
525 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
527 if (sbinfo->free_inodes >= 0) {
528 spin_lock(&sbinfo->stat_lock);
529 if (unlikely(!sbinfo->free_inodes)) {
530 spin_unlock(&sbinfo->stat_lock);
533 sbinfo->free_inodes--;
534 spin_unlock(&sbinfo->stat_lock);
540 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
542 if (sbinfo->free_inodes >= 0) {
543 spin_lock(&sbinfo->stat_lock);
544 sbinfo->free_inodes++;
545 spin_unlock(&sbinfo->stat_lock);
550 static struct kmem_cache *hugetlbfs_inode_cachep;
552 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
554 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
555 struct hugetlbfs_inode_info *p;
557 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
559 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
561 hugetlbfs_inc_free_inodes(sbinfo);
564 return &p->vfs_inode;
567 static void hugetlbfs_destroy_inode(struct inode *inode)
569 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
570 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
571 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
574 static const struct address_space_operations hugetlbfs_aops = {
575 .readpage = hugetlbfs_readpage,
576 .write_begin = hugetlbfs_write_begin,
577 .write_end = hugetlbfs_write_end,
578 .set_page_dirty = hugetlbfs_set_page_dirty,
582 static void init_once(void *foo, struct kmem_cache *cachep, unsigned long flags)
584 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
586 inode_init_once(&ei->vfs_inode);
589 const struct file_operations hugetlbfs_file_operations = {
590 .mmap = hugetlbfs_file_mmap,
591 .fsync = simple_sync_file,
592 .get_unmapped_area = hugetlb_get_unmapped_area,
595 static const struct inode_operations hugetlbfs_dir_inode_operations = {
596 .create = hugetlbfs_create,
597 .lookup = simple_lookup,
599 .unlink = simple_unlink,
600 .symlink = hugetlbfs_symlink,
601 .mkdir = hugetlbfs_mkdir,
602 .rmdir = simple_rmdir,
603 .mknod = hugetlbfs_mknod,
604 .rename = simple_rename,
605 .setattr = hugetlbfs_setattr,
608 static const struct inode_operations hugetlbfs_inode_operations = {
609 .setattr = hugetlbfs_setattr,
612 static const struct super_operations hugetlbfs_ops = {
613 .alloc_inode = hugetlbfs_alloc_inode,
614 .destroy_inode = hugetlbfs_destroy_inode,
615 .statfs = hugetlbfs_statfs,
616 .delete_inode = hugetlbfs_delete_inode,
617 .drop_inode = hugetlbfs_drop_inode,
618 .put_super = hugetlbfs_put_super,
622 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
625 substring_t args[MAX_OPT_ARGS];
631 while ((p = strsep(&options, ",")) != NULL) {
636 token = match_token(p, tokens, args);
639 if (match_int(&args[0], &option))
641 pconfig->uid = option;
645 if (match_int(&args[0], &option))
647 pconfig->gid = option;
651 if (match_octal(&args[0], &option))
653 pconfig->mode = option & 0777U;
657 unsigned long long size;
658 /* memparse() will accept a K/M/G without a digit */
659 if (!isdigit(*args[0].from))
661 size = memparse(args[0].from, &rest);
663 size <<= HPAGE_SHIFT;
664 size *= max_huge_pages;
667 pconfig->nr_blocks = (size >> HPAGE_SHIFT);
672 /* memparse() will accept a K/M/G without a digit */
673 if (!isdigit(*args[0].from))
675 pconfig->nr_inodes = memparse(args[0].from, &rest);
679 printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
688 printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
694 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
696 struct inode * inode;
697 struct dentry * root;
699 struct hugetlbfs_config config;
700 struct hugetlbfs_sb_info *sbinfo;
702 config.nr_blocks = -1; /* No limit on size by default */
703 config.nr_inodes = -1; /* No limit on number of inodes by default */
704 config.uid = current->fsuid;
705 config.gid = current->fsgid;
707 ret = hugetlbfs_parse_options(data, &config);
711 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
714 sb->s_fs_info = sbinfo;
715 spin_lock_init(&sbinfo->stat_lock);
716 sbinfo->max_blocks = config.nr_blocks;
717 sbinfo->free_blocks = config.nr_blocks;
718 sbinfo->max_inodes = config.nr_inodes;
719 sbinfo->free_inodes = config.nr_inodes;
720 sb->s_maxbytes = MAX_LFS_FILESIZE;
721 sb->s_blocksize = HPAGE_SIZE;
722 sb->s_blocksize_bits = HPAGE_SHIFT;
723 sb->s_magic = HUGETLBFS_MAGIC;
724 sb->s_op = &hugetlbfs_ops;
726 inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
727 S_IFDIR | config.mode, 0);
731 root = d_alloc_root(inode);
743 int hugetlb_get_quota(struct address_space *mapping)
746 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
748 if (sbinfo->free_blocks > -1) {
749 spin_lock(&sbinfo->stat_lock);
750 if (sbinfo->free_blocks > 0)
751 sbinfo->free_blocks--;
754 spin_unlock(&sbinfo->stat_lock);
760 void hugetlb_put_quota(struct address_space *mapping)
762 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
764 if (sbinfo->free_blocks > -1) {
765 spin_lock(&sbinfo->stat_lock);
766 sbinfo->free_blocks++;
767 spin_unlock(&sbinfo->stat_lock);
771 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
772 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
774 return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
777 static struct file_system_type hugetlbfs_fs_type = {
779 .get_sb = hugetlbfs_get_sb,
780 .kill_sb = kill_litter_super,
783 static struct vfsmount *hugetlbfs_vfsmount;
785 static int can_do_hugetlb_shm(void)
787 return likely(capable(CAP_IPC_LOCK) ||
788 in_group_p(sysctl_hugetlb_shm_group) ||
792 struct file *hugetlb_file_setup(const char *name, size_t size)
797 struct dentry *dentry, *root;
798 struct qstr quick_string;
800 if (!hugetlbfs_vfsmount)
801 return ERR_PTR(-ENOENT);
803 if (!can_do_hugetlb_shm())
804 return ERR_PTR(-EPERM);
806 if (!user_shm_lock(size, current->user))
807 return ERR_PTR(-ENOMEM);
809 root = hugetlbfs_vfsmount->mnt_root;
810 quick_string.name = name;
811 quick_string.len = strlen(quick_string.name);
812 quick_string.hash = 0;
813 dentry = d_alloc(root, &quick_string);
818 file = get_empty_filp();
823 inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
824 current->fsgid, S_IFREG | S_IRWXUGO, 0);
829 if (hugetlb_reserve_pages(inode, 0, size >> HPAGE_SHIFT))
832 d_instantiate(dentry, inode);
833 inode->i_size = size;
835 file->f_path.mnt = mntget(hugetlbfs_vfsmount);
836 file->f_path.dentry = dentry;
837 file->f_mapping = inode->i_mapping;
838 file->f_op = &hugetlbfs_file_operations;
839 file->f_mode = FMODE_WRITE | FMODE_READ;
849 user_shm_unlock(size, current->user);
850 return ERR_PTR(error);
853 static int __init init_hugetlbfs_fs(void)
856 struct vfsmount *vfsmount;
858 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
859 sizeof(struct hugetlbfs_inode_info),
861 if (hugetlbfs_inode_cachep == NULL)
864 error = register_filesystem(&hugetlbfs_fs_type);
868 vfsmount = kern_mount(&hugetlbfs_fs_type);
870 if (!IS_ERR(vfsmount)) {
871 hugetlbfs_vfsmount = vfsmount;
875 error = PTR_ERR(vfsmount);
879 kmem_cache_destroy(hugetlbfs_inode_cachep);
883 static void __exit exit_hugetlbfs_fs(void)
885 kmem_cache_destroy(hugetlbfs_inode_cachep);
886 unregister_filesystem(&hugetlbfs_fs_type);
889 module_init(init_hugetlbfs_fs)
890 module_exit(exit_hugetlbfs_fs)
892 MODULE_LICENSE("GPL");