1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (c) 2013, Intel Corporation
5 * Authors: Huajun Li <huajun.li@intel.com>
6 * Haicheng Li <haicheng.li@intel.com>
10 #include <linux/f2fs_fs.h>
11 #include <linux/fiemap.h>
15 #include <trace/events/f2fs.h>
17 static bool support_inline_data(struct inode *inode)
19 if (f2fs_used_in_atomic_write(inode))
21 if (!S_ISREG(inode->i_mode) && !S_ISLNK(inode->i_mode))
23 if (i_size_read(inode) > MAX_INLINE_DATA(inode))
28 bool f2fs_may_inline_data(struct inode *inode)
30 if (!support_inline_data(inode))
33 return !f2fs_post_read_required(inode);
36 static bool inode_has_blocks(struct inode *inode, struct page *ipage)
38 struct f2fs_inode *ri = F2FS_INODE(ipage);
41 if (F2FS_HAS_BLOCKS(inode))
44 for (i = 0; i < DEF_NIDS_PER_INODE; i++) {
51 bool f2fs_sanity_check_inline_data(struct inode *inode, struct page *ipage)
53 if (!f2fs_has_inline_data(inode))
56 if (inode_has_blocks(inode, ipage))
59 if (!support_inline_data(inode))
63 * used by sanity_check_inode(), when disk layout fields has not
64 * been synchronized to inmem fields.
66 return (S_ISREG(inode->i_mode) &&
67 (file_is_encrypt(inode) || file_is_verity(inode) ||
68 (F2FS_I(inode)->i_flags & F2FS_COMPR_FL)));
71 bool f2fs_may_inline_dentry(struct inode *inode)
73 if (!test_opt(F2FS_I_SB(inode), INLINE_DENTRY))
76 if (!S_ISDIR(inode->i_mode))
82 void f2fs_do_read_inline_data(struct folio *folio, struct folio *ifolio)
84 struct inode *inode = folio->mapping->host;
86 if (folio_test_uptodate(folio))
89 f2fs_bug_on(F2FS_I_SB(inode), folio->index);
91 folio_zero_segment(folio, MAX_INLINE_DATA(inode), folio_size(folio));
93 /* Copy the whole inline data block */
94 memcpy_to_folio(folio, 0, inline_data_addr(inode, ifolio),
95 MAX_INLINE_DATA(inode));
96 if (!folio_test_uptodate(folio))
97 folio_mark_uptodate(folio);
100 void f2fs_truncate_inline_inode(struct inode *inode, struct folio *ifolio,
105 if (from >= MAX_INLINE_DATA(inode))
108 addr = inline_data_addr(inode, ifolio);
110 f2fs_folio_wait_writeback(ifolio, NODE, true, true);
111 memset(addr + from, 0, MAX_INLINE_DATA(inode) - from);
112 folio_mark_dirty(ifolio);
115 clear_inode_flag(inode, FI_DATA_EXIST);
118 int f2fs_read_inline_data(struct inode *inode, struct folio *folio)
120 struct folio *ifolio;
122 ifolio = f2fs_get_inode_folio(F2FS_I_SB(inode), inode->i_ino);
123 if (IS_ERR(ifolio)) {
125 return PTR_ERR(ifolio);
128 if (!f2fs_has_inline_data(inode)) {
129 f2fs_folio_put(ifolio, true);
134 folio_zero_segment(folio, 0, folio_size(folio));
136 f2fs_do_read_inline_data(folio, ifolio);
138 if (!folio_test_uptodate(folio))
139 folio_mark_uptodate(folio);
140 f2fs_folio_put(ifolio, true);
145 int f2fs_convert_inline_folio(struct dnode_of_data *dn, struct folio *folio)
147 struct f2fs_io_info fio = {
148 .sbi = F2FS_I_SB(dn->inode),
149 .ino = dn->inode->i_ino,
152 .op_flags = REQ_SYNC | REQ_PRIO,
153 .page = &folio->page,
154 .encrypted_page = NULL,
155 .io_type = FS_DATA_IO,
160 if (!f2fs_exist_data(dn->inode))
163 err = f2fs_reserve_block(dn, 0);
167 err = f2fs_get_node_info(fio.sbi, dn->nid, &ni, false);
169 f2fs_truncate_data_blocks_range(dn, 1);
174 fio.version = ni.version;
176 if (unlikely(dn->data_blkaddr != NEW_ADDR)) {
178 set_sbi_flag(fio.sbi, SBI_NEED_FSCK);
179 f2fs_warn(fio.sbi, "%s: corrupted inline inode ino=%lx, i_addr[0]:0x%x, run fsck to fix.",
180 __func__, dn->inode->i_ino, dn->data_blkaddr);
181 f2fs_handle_error(fio.sbi, ERROR_INVALID_BLKADDR);
182 return -EFSCORRUPTED;
185 f2fs_bug_on(F2FS_F_SB(folio), folio_test_writeback(folio));
187 f2fs_do_read_inline_data(folio, dn->inode_folio);
188 folio_mark_dirty(folio);
190 /* clear dirty state */
191 dirty = folio_clear_dirty_for_io(folio);
193 /* write data page to try to make data consistent */
194 folio_start_writeback(folio);
195 fio.old_blkaddr = dn->data_blkaddr;
196 set_inode_flag(dn->inode, FI_HOT_DATA);
197 f2fs_outplace_write_data(dn, &fio);
198 f2fs_folio_wait_writeback(folio, DATA, true, true);
200 inode_dec_dirty_pages(dn->inode);
201 f2fs_remove_dirty_inode(dn->inode);
204 /* this converted inline_data should be recovered. */
205 set_inode_flag(dn->inode, FI_APPEND_WRITE);
207 /* clear inline data and flag after data writeback */
208 f2fs_truncate_inline_inode(dn->inode, dn->inode_folio, 0);
209 clear_page_private_inline(&dn->inode_folio->page);
211 stat_dec_inline_inode(dn->inode);
212 clear_inode_flag(dn->inode, FI_INLINE_DATA);
217 int f2fs_convert_inline_inode(struct inode *inode)
219 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
220 struct dnode_of_data dn;
221 struct folio *ifolio, *folio;
224 if (f2fs_hw_is_readonly(sbi) || f2fs_readonly(sbi->sb))
227 if (!f2fs_has_inline_data(inode))
230 err = f2fs_dquot_initialize(inode);
234 folio = f2fs_grab_cache_folio(inode->i_mapping, 0, false);
236 return PTR_ERR(folio);
240 ifolio = f2fs_get_inode_folio(sbi, inode->i_ino);
241 if (IS_ERR(ifolio)) {
242 err = PTR_ERR(ifolio);
246 set_new_dnode(&dn, inode, ifolio, ifolio, 0);
248 if (f2fs_has_inline_data(inode))
249 err = f2fs_convert_inline_folio(&dn, folio);
255 f2fs_folio_put(folio, true);
258 f2fs_balance_fs(sbi, dn.node_changed);
263 int f2fs_write_inline_data(struct inode *inode, struct folio *folio)
265 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
266 struct folio *ifolio;
268 ifolio = f2fs_get_inode_folio(sbi, inode->i_ino);
270 return PTR_ERR(ifolio);
272 if (!f2fs_has_inline_data(inode)) {
273 f2fs_folio_put(ifolio, true);
277 f2fs_bug_on(F2FS_I_SB(inode), folio->index);
279 f2fs_folio_wait_writeback(ifolio, NODE, true, true);
280 memcpy_from_folio(inline_data_addr(inode, ifolio),
281 folio, 0, MAX_INLINE_DATA(inode));
282 folio_mark_dirty(ifolio);
284 f2fs_clear_page_cache_dirty_tag(folio);
286 set_inode_flag(inode, FI_APPEND_WRITE);
287 set_inode_flag(inode, FI_DATA_EXIST);
289 clear_page_private_inline(&ifolio->page);
290 f2fs_folio_put(ifolio, 1);
294 int f2fs_recover_inline_data(struct inode *inode, struct folio *nfolio)
296 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
297 struct f2fs_inode *ri = NULL;
298 void *src_addr, *dst_addr;
301 * The inline_data recovery policy is as follows.
302 * [prev.] [next] of inline_data flag
303 * o o -> recover inline_data
304 * o x -> remove inline_data, and then recover data blocks
305 * x o -> remove data blocks, and then recover inline_data
306 * x x -> recover data blocks
308 if (IS_INODE(&nfolio->page))
309 ri = F2FS_INODE(&nfolio->page);
311 if (f2fs_has_inline_data(inode) &&
312 ri && (ri->i_inline & F2FS_INLINE_DATA)) {
313 struct folio *ifolio;
315 ifolio = f2fs_get_inode_folio(sbi, inode->i_ino);
317 return PTR_ERR(ifolio);
319 f2fs_folio_wait_writeback(ifolio, NODE, true, true);
321 src_addr = inline_data_addr(inode, nfolio);
322 dst_addr = inline_data_addr(inode, ifolio);
323 memcpy(dst_addr, src_addr, MAX_INLINE_DATA(inode));
325 set_inode_flag(inode, FI_INLINE_DATA);
326 set_inode_flag(inode, FI_DATA_EXIST);
328 folio_mark_dirty(ifolio);
329 f2fs_folio_put(ifolio, true);
333 if (f2fs_has_inline_data(inode)) {
334 struct folio *ifolio = f2fs_get_inode_folio(sbi, inode->i_ino);
336 return PTR_ERR(ifolio);
337 f2fs_truncate_inline_inode(inode, ifolio, 0);
338 stat_dec_inline_inode(inode);
339 clear_inode_flag(inode, FI_INLINE_DATA);
340 f2fs_folio_put(ifolio, true);
341 } else if (ri && (ri->i_inline & F2FS_INLINE_DATA)) {
344 ret = f2fs_truncate_blocks(inode, 0, false);
347 stat_inc_inline_inode(inode);
353 struct f2fs_dir_entry *f2fs_find_in_inline_dir(struct inode *dir,
354 const struct f2fs_filename *fname,
355 struct folio **res_folio,
358 struct f2fs_sb_info *sbi = F2FS_SB(dir->i_sb);
359 struct f2fs_dir_entry *de;
360 struct f2fs_dentry_ptr d;
361 struct folio *ifolio;
364 ifolio = f2fs_get_inode_folio(sbi, dir->i_ino);
365 if (IS_ERR(ifolio)) {
370 inline_dentry = inline_data_addr(dir, ifolio);
372 make_dentry_ptr_inline(dir, &d, inline_dentry);
373 de = f2fs_find_target_dentry(&d, fname, NULL, use_hash);
374 folio_unlock(ifolio);
376 *res_folio = ERR_CAST(de);
382 f2fs_folio_put(ifolio, false);
387 int f2fs_make_empty_inline_dir(struct inode *inode, struct inode *parent,
388 struct folio *ifolio)
390 struct f2fs_dentry_ptr d;
393 inline_dentry = inline_data_addr(inode, ifolio);
395 make_dentry_ptr_inline(inode, &d, inline_dentry);
396 f2fs_do_make_empty_dir(inode, parent, &d);
398 folio_mark_dirty(ifolio);
400 /* update i_size to MAX_INLINE_DATA */
401 if (i_size_read(inode) < MAX_INLINE_DATA(inode))
402 f2fs_i_size_write(inode, MAX_INLINE_DATA(inode));
407 * NOTE: ipage is grabbed by caller, but if any error occurs, we should
408 * release ipage in this function.
410 static int f2fs_move_inline_dirents(struct inode *dir, struct folio *ifolio,
414 struct dnode_of_data dn;
415 struct f2fs_dentry_block *dentry_blk;
416 struct f2fs_dentry_ptr src, dst;
419 folio = f2fs_grab_cache_folio(dir->i_mapping, 0, true);
421 f2fs_folio_put(ifolio, true);
422 return PTR_ERR(folio);
425 set_new_dnode(&dn, dir, ifolio, NULL, 0);
426 err = f2fs_reserve_block(&dn, 0);
430 if (unlikely(dn.data_blkaddr != NEW_ADDR)) {
432 set_sbi_flag(F2FS_F_SB(folio), SBI_NEED_FSCK);
433 f2fs_warn(F2FS_F_SB(folio), "%s: corrupted inline inode ino=%lx, i_addr[0]:0x%x, run fsck to fix.",
434 __func__, dir->i_ino, dn.data_blkaddr);
435 f2fs_handle_error(F2FS_F_SB(folio), ERROR_INVALID_BLKADDR);
440 f2fs_folio_wait_writeback(folio, DATA, true, true);
442 dentry_blk = folio_address(folio);
445 * Start by zeroing the full block, to ensure that all unused space is
446 * zeroed and no uninitialized memory is leaked to disk.
448 memset(dentry_blk, 0, F2FS_BLKSIZE);
450 make_dentry_ptr_inline(dir, &src, inline_dentry);
451 make_dentry_ptr_block(dir, &dst, dentry_blk);
453 /* copy data from inline dentry block to new dentry block */
454 memcpy(dst.bitmap, src.bitmap, src.nr_bitmap);
455 memcpy(dst.dentry, src.dentry, SIZE_OF_DIR_ENTRY * src.max);
456 memcpy(dst.filename, src.filename, src.max * F2FS_SLOT_LEN);
458 if (!folio_test_uptodate(folio))
459 folio_mark_uptodate(folio);
460 folio_mark_dirty(folio);
462 /* clear inline dir and flag after data writeback */
463 f2fs_truncate_inline_inode(dir, ifolio, 0);
465 stat_dec_inline_dir(dir);
466 clear_inode_flag(dir, FI_INLINE_DENTRY);
469 * should retrieve reserved space which was used to keep
470 * inline_dentry's structure for backward compatibility.
472 if (!f2fs_sb_has_flexible_inline_xattr(F2FS_I_SB(dir)) &&
473 !f2fs_has_inline_xattr(dir))
474 F2FS_I(dir)->i_inline_xattr_size = 0;
476 f2fs_i_depth_write(dir, 1);
477 if (i_size_read(dir) < PAGE_SIZE)
478 f2fs_i_size_write(dir, PAGE_SIZE);
480 f2fs_folio_put(folio, true);
484 static int f2fs_add_inline_entries(struct inode *dir, void *inline_dentry)
486 struct f2fs_dentry_ptr d;
487 unsigned long bit_pos = 0;
490 make_dentry_ptr_inline(dir, &d, inline_dentry);
492 while (bit_pos < d.max) {
493 struct f2fs_dir_entry *de;
494 struct f2fs_filename fname;
498 if (!test_bit_le(bit_pos, d.bitmap)) {
503 de = &d.dentry[bit_pos];
505 if (unlikely(!de->name_len)) {
511 * We only need the disk_name and hash to move the dentry.
512 * We don't need the original or casefolded filenames.
514 memset(&fname, 0, sizeof(fname));
515 fname.disk_name.name = d.filename[bit_pos];
516 fname.disk_name.len = le16_to_cpu(de->name_len);
517 fname.hash = de->hash_code;
519 ino = le32_to_cpu(de->ino);
520 fake_mode = fs_ftype_to_dtype(de->file_type) << S_DT_SHIFT;
522 err = f2fs_add_regular_entry(dir, &fname, NULL, ino, fake_mode);
524 goto punch_dentry_pages;
526 bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
530 truncate_inode_pages(&dir->i_data, 0);
531 f2fs_truncate_blocks(dir, 0, false);
532 f2fs_remove_dirty_inode(dir);
536 static int f2fs_move_rehashed_dirents(struct inode *dir, struct folio *ifolio,
542 backup_dentry = f2fs_kmalloc(F2FS_I_SB(dir),
543 MAX_INLINE_DATA(dir), GFP_F2FS_ZERO);
544 if (!backup_dentry) {
545 f2fs_folio_put(ifolio, true);
549 memcpy(backup_dentry, inline_dentry, MAX_INLINE_DATA(dir));
550 f2fs_truncate_inline_inode(dir, ifolio, 0);
552 folio_unlock(ifolio);
554 err = f2fs_add_inline_entries(dir, backup_dentry);
560 stat_dec_inline_dir(dir);
561 clear_inode_flag(dir, FI_INLINE_DENTRY);
564 * should retrieve reserved space which was used to keep
565 * inline_dentry's structure for backward compatibility.
567 if (!f2fs_sb_has_flexible_inline_xattr(F2FS_I_SB(dir)) &&
568 !f2fs_has_inline_xattr(dir))
569 F2FS_I(dir)->i_inline_xattr_size = 0;
571 kfree(backup_dentry);
575 f2fs_folio_wait_writeback(ifolio, NODE, true, true);
576 memcpy(inline_dentry, backup_dentry, MAX_INLINE_DATA(dir));
577 f2fs_i_depth_write(dir, 0);
578 f2fs_i_size_write(dir, MAX_INLINE_DATA(dir));
579 folio_mark_dirty(ifolio);
580 f2fs_folio_put(ifolio, 1);
582 kfree(backup_dentry);
586 static int do_convert_inline_dir(struct inode *dir, struct folio *ifolio,
589 if (!F2FS_I(dir)->i_dir_level)
590 return f2fs_move_inline_dirents(dir, ifolio, inline_dentry);
592 return f2fs_move_rehashed_dirents(dir, ifolio, inline_dentry);
595 int f2fs_try_convert_inline_dir(struct inode *dir, struct dentry *dentry)
597 struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
598 struct folio *ifolio;
599 struct f2fs_filename fname;
600 void *inline_dentry = NULL;
603 if (!f2fs_has_inline_dentry(dir))
608 err = f2fs_setup_filename(dir, &dentry->d_name, 0, &fname);
612 ifolio = f2fs_get_inode_folio(sbi, dir->i_ino);
613 if (IS_ERR(ifolio)) {
614 err = PTR_ERR(ifolio);
618 if (f2fs_has_enough_room(dir, ifolio, &fname)) {
619 f2fs_folio_put(ifolio, true);
623 inline_dentry = inline_data_addr(dir, ifolio);
625 err = do_convert_inline_dir(dir, ifolio, inline_dentry);
627 f2fs_folio_put(ifolio, true);
629 f2fs_free_filename(&fname);
635 int f2fs_add_inline_entry(struct inode *dir, const struct f2fs_filename *fname,
636 struct inode *inode, nid_t ino, umode_t mode)
638 struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
639 struct folio *ifolio;
640 unsigned int bit_pos;
641 void *inline_dentry = NULL;
642 struct f2fs_dentry_ptr d;
643 int slots = GET_DENTRY_SLOTS(fname->disk_name.len);
644 struct folio *folio = NULL;
647 ifolio = f2fs_get_inode_folio(sbi, dir->i_ino);
649 return PTR_ERR(ifolio);
651 inline_dentry = inline_data_addr(dir, ifolio);
652 make_dentry_ptr_inline(dir, &d, inline_dentry);
654 bit_pos = f2fs_room_for_filename(d.bitmap, slots, d.max);
655 if (bit_pos >= d.max) {
656 err = do_convert_inline_dir(dir, ifolio, inline_dentry);
664 f2fs_down_write_nested(&F2FS_I(inode)->i_sem,
665 SINGLE_DEPTH_NESTING);
666 folio = f2fs_init_inode_metadata(inode, dir, fname, ifolio);
668 err = PTR_ERR(folio);
673 f2fs_folio_wait_writeback(ifolio, NODE, true, true);
675 f2fs_update_dentry(ino, mode, &d, &fname->disk_name, fname->hash,
678 folio_mark_dirty(ifolio);
680 /* we don't need to mark_inode_dirty now */
682 f2fs_i_pino_write(inode, dir->i_ino);
684 /* synchronize inode page's data from inode cache */
685 if (is_inode_flag_set(inode, FI_NEW_INODE))
686 f2fs_update_inode(inode, folio);
688 f2fs_folio_put(folio, true);
691 f2fs_update_parent_metadata(dir, inode, 0);
694 f2fs_up_write(&F2FS_I(inode)->i_sem);
696 f2fs_folio_put(ifolio, true);
700 void f2fs_delete_inline_entry(struct f2fs_dir_entry *dentry,
701 struct folio *folio, struct inode *dir, struct inode *inode)
703 struct f2fs_dentry_ptr d;
705 int slots = GET_DENTRY_SLOTS(le16_to_cpu(dentry->name_len));
706 unsigned int bit_pos;
710 f2fs_folio_wait_writeback(folio, NODE, true, true);
712 inline_dentry = inline_data_addr(dir, folio);
713 make_dentry_ptr_inline(dir, &d, inline_dentry);
715 bit_pos = dentry - d.dentry;
716 for (i = 0; i < slots; i++)
717 __clear_bit_le(bit_pos + i, d.bitmap);
719 folio_mark_dirty(folio);
720 f2fs_folio_put(folio, true);
722 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
723 f2fs_mark_inode_dirty_sync(dir, false);
726 f2fs_drop_nlink(dir, inode);
729 bool f2fs_empty_inline_dir(struct inode *dir)
731 struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
732 struct folio *ifolio;
733 unsigned int bit_pos = 2;
735 struct f2fs_dentry_ptr d;
737 ifolio = f2fs_get_inode_folio(sbi, dir->i_ino);
741 inline_dentry = inline_data_addr(dir, ifolio);
742 make_dentry_ptr_inline(dir, &d, inline_dentry);
744 bit_pos = find_next_bit_le(d.bitmap, d.max, bit_pos);
746 f2fs_folio_put(ifolio, true);
754 int f2fs_read_inline_dir(struct file *file, struct dir_context *ctx,
755 struct fscrypt_str *fstr)
757 struct inode *inode = file_inode(file);
758 struct folio *ifolio = NULL;
759 struct f2fs_dentry_ptr d;
760 void *inline_dentry = NULL;
763 make_dentry_ptr_inline(inode, &d, inline_dentry);
765 if (ctx->pos == d.max)
768 ifolio = f2fs_get_inode_folio(F2FS_I_SB(inode), inode->i_ino);
770 return PTR_ERR(ifolio);
773 * f2fs_readdir was protected by inode.i_rwsem, it is safe to access
774 * ipage without page's lock held.
776 folio_unlock(ifolio);
778 inline_dentry = inline_data_addr(inode, ifolio);
780 make_dentry_ptr_inline(inode, &d, inline_dentry);
782 err = f2fs_fill_dentries(ctx, &d, 0, fstr);
786 f2fs_folio_put(ifolio, false);
787 return err < 0 ? err : 0;
790 int f2fs_inline_data_fiemap(struct inode *inode,
791 struct fiemap_extent_info *fieinfo, __u64 start, __u64 len)
793 __u64 byteaddr, ilen;
794 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED |
797 struct folio *ifolio;
800 ifolio = f2fs_get_inode_folio(F2FS_I_SB(inode), inode->i_ino);
802 return PTR_ERR(ifolio);
804 if ((S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
805 !f2fs_has_inline_data(inode)) {
810 if (S_ISDIR(inode->i_mode) && !f2fs_has_inline_dentry(inode)) {
815 ilen = min_t(size_t, MAX_INLINE_DATA(inode), i_size_read(inode));
818 if (start + len < ilen)
822 err = f2fs_get_node_info(F2FS_I_SB(inode), inode->i_ino, &ni, false);
826 byteaddr = (__u64)ni.blk_addr << inode->i_sb->s_blocksize_bits;
827 byteaddr += (char *)inline_data_addr(inode, ifolio) -
828 (char *)F2FS_INODE(&ifolio->page);
829 err = fiemap_fill_next_extent(fieinfo, start, byteaddr, ilen, flags);
830 trace_f2fs_fiemap(inode, start, byteaddr, ilen, flags, err);
832 f2fs_folio_put(ifolio, true);