4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
17 static struct kmem_cache *fsync_entry_slab;
19 bool space_for_roll_forward(struct f2fs_sb_info *sbi)
21 if (sbi->last_valid_block_count + sbi->alloc_valid_block_count
22 > sbi->user_block_count)
27 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
30 struct list_head *this;
31 struct fsync_inode_entry *entry;
33 list_for_each(this, head) {
34 entry = list_entry(this, struct fsync_inode_entry, list);
35 if (entry->inode->i_ino == ino)
41 static int recover_dentry(struct page *ipage, struct inode *inode)
43 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
44 nid_t pino = le32_to_cpu(raw_inode->i_pino);
45 struct f2fs_dir_entry *de;
48 struct inode *dir, *einode;
51 dir = check_dirty_dir_inode(F2FS_SB(inode->i_sb), pino);
53 dir = f2fs_iget(inode->i_sb, pino);
58 set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
59 add_dirty_dir_inode(dir);
62 name.len = le32_to_cpu(raw_inode->i_namelen);
63 name.name = raw_inode->i_name;
65 if (unlikely(name.len > F2FS_NAME_LEN)) {
71 de = f2fs_find_entry(dir, &name, &page);
72 if (de && inode->i_ino == le32_to_cpu(de->ino))
75 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
78 if (PTR_ERR(einode) == -ENOENT)
82 err = acquire_orphan_inode(F2FS_SB(inode->i_sb));
87 f2fs_delete_entry(de, page, einode);
91 err = __f2fs_add_link(dir, &name, inode);
96 f2fs_put_page(page, 0);
98 f2fs_msg(inode->i_sb, KERN_NOTICE,
99 "%s: ino = %x, name = %s, dir = %lx, err = %d",
100 __func__, ino_of_node(ipage), raw_inode->i_name,
101 IS_ERR(dir) ? 0 : dir->i_ino, err);
105 static int recover_inode(struct inode *inode, struct page *node_page)
107 struct f2fs_inode *raw_inode = F2FS_INODE(node_page);
109 if (!IS_INODE(node_page))
112 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
113 i_size_write(inode, le64_to_cpu(raw_inode->i_size));
114 inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
115 inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
116 inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
117 inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
118 inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
119 inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
121 if (is_dent_dnode(node_page))
122 return recover_dentry(node_page, inode);
124 f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
125 ino_of_node(node_page), raw_inode->i_name);
129 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
131 unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
132 struct curseg_info *curseg;
137 /* get node pages in the current segment */
138 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
139 blkaddr = START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff;
142 page = alloc_page(GFP_F2FS_ZERO);
148 struct fsync_inode_entry *entry;
150 err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC);
156 if (cp_ver != cpver_of_node(page))
159 if (!is_fsync_dnode(page))
162 entry = get_fsync_inode(head, ino_of_node(page));
164 if (IS_INODE(page) && is_dent_dnode(page))
165 set_inode_flag(F2FS_I(entry->inode),
168 if (IS_INODE(page) && is_dent_dnode(page)) {
169 err = recover_inode_page(sbi, page);
174 /* add this fsync inode to the list */
175 entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
181 entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
182 if (IS_ERR(entry->inode)) {
183 err = PTR_ERR(entry->inode);
184 kmem_cache_free(fsync_entry_slab, entry);
187 list_add_tail(&entry->list, head);
189 entry->blkaddr = blkaddr;
191 err = recover_inode(entry->inode, page);
192 if (err && err != -ENOENT)
195 /* check next segment */
196 blkaddr = next_blkaddr_of_node(page);
200 __free_pages(page, 0);
205 static void destroy_fsync_dnodes(struct list_head *head)
207 struct fsync_inode_entry *entry, *tmp;
209 list_for_each_entry_safe(entry, tmp, head, list) {
211 list_del(&entry->list);
212 kmem_cache_free(fsync_entry_slab, entry);
216 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
217 block_t blkaddr, struct dnode_of_data *dn)
219 struct seg_entry *sentry;
220 unsigned int segno = GET_SEGNO(sbi, blkaddr);
221 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
222 struct f2fs_summary sum;
226 struct page *node_page;
231 sentry = get_seg_entry(sbi, segno);
232 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
235 /* Get the previous summary */
236 for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
237 struct curseg_info *curseg = CURSEG_I(sbi, i);
238 if (curseg->segno == segno) {
239 sum = curseg->sum_blk->entries[blkoff];
243 if (i > CURSEG_COLD_DATA) {
244 struct page *sum_page = get_sum_page(sbi, segno);
245 struct f2fs_summary_block *sum_node;
246 kaddr = page_address(sum_page);
247 sum_node = (struct f2fs_summary_block *)kaddr;
248 sum = sum_node->entries[blkoff];
249 f2fs_put_page(sum_page, 1);
252 /* Use the locked dnode page and inode */
253 nid = le32_to_cpu(sum.nid);
254 if (dn->inode->i_ino == nid) {
255 struct dnode_of_data tdn = *dn;
257 tdn.node_page = dn->inode_page;
258 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
259 truncate_data_blocks_range(&tdn, 1);
261 } else if (dn->nid == nid) {
262 struct dnode_of_data tdn = *dn;
263 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
264 truncate_data_blocks_range(&tdn, 1);
268 /* Get the node page */
269 node_page = get_node_page(sbi, nid);
270 if (IS_ERR(node_page))
271 return PTR_ERR(node_page);
273 offset = ofs_of_node(node_page);
274 ino = ino_of_node(node_page);
275 f2fs_put_page(node_page, 1);
277 /* Deallocate previous index in the node page */
278 inode = f2fs_iget(sbi->sb, ino);
280 return PTR_ERR(inode);
282 bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
283 le16_to_cpu(sum.ofs_in_node);
285 truncate_hole(inode, bidx, bidx + 1);
290 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
291 struct page *page, block_t blkaddr)
293 struct f2fs_inode_info *fi = F2FS_I(inode);
294 unsigned int start, end;
295 struct dnode_of_data dn;
296 struct f2fs_summary sum;
298 int err = 0, recovered = 0;
300 if (recover_inline_data(inode, page))
303 if (recover_xattr_data(inode, page, blkaddr))
306 start = start_bidx_of_node(ofs_of_node(page), fi);
308 end = start + ADDRS_PER_INODE(fi);
310 end = start + ADDRS_PER_BLOCK;
314 set_new_dnode(&dn, inode, NULL, NULL, 0);
316 err = get_dnode_of_data(&dn, start, ALLOC_NODE);
322 wait_on_page_writeback(dn.node_page);
324 get_node_info(sbi, dn.nid, &ni);
325 f2fs_bug_on(ni.ino != ino_of_node(page));
326 f2fs_bug_on(ofs_of_node(dn.node_page) != ofs_of_node(page));
328 for (; start < end; start++) {
331 src = datablock_addr(dn.node_page, dn.ofs_in_node);
332 dest = datablock_addr(page, dn.ofs_in_node);
334 if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
335 if (src == NULL_ADDR) {
336 err = reserve_new_block(&dn);
337 /* We should not get -ENOSPC */
341 /* Check the previous node page having this index */
342 err = check_index_in_prev_nodes(sbi, dest, &dn);
346 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
348 /* write dummy data page */
349 recover_data_page(sbi, NULL, &sum, src, dest);
350 update_extent_cache(dest, &dn);
356 /* write node page in place */
357 set_summary(&sum, dn.nid, 0, 0);
358 if (IS_INODE(dn.node_page))
359 sync_inode_page(&dn);
361 copy_node_footer(dn.node_page, page);
362 fill_node_footer(dn.node_page, dn.nid, ni.ino,
363 ofs_of_node(page), false);
364 set_page_dirty(dn.node_page);
366 recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr);
371 f2fs_msg(sbi->sb, KERN_NOTICE,
372 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
373 inode->i_ino, recovered, err);
377 static int recover_data(struct f2fs_sb_info *sbi,
378 struct list_head *head, int type)
380 unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
381 struct curseg_info *curseg;
386 /* get node pages in the current segment */
387 curseg = CURSEG_I(sbi, type);
388 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
391 page = alloc_page(GFP_F2FS_ZERO);
398 struct fsync_inode_entry *entry;
400 err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC);
406 if (cp_ver != cpver_of_node(page))
409 entry = get_fsync_inode(head, ino_of_node(page));
413 err = do_recover_data(sbi, entry->inode, page, blkaddr);
417 if (entry->blkaddr == blkaddr) {
419 list_del(&entry->list);
420 kmem_cache_free(fsync_entry_slab, entry);
423 /* check next segment */
424 blkaddr = next_blkaddr_of_node(page);
428 __free_pages(page, 0);
431 allocate_new_segments(sbi);
435 int recover_fsync_data(struct f2fs_sb_info *sbi)
437 struct list_head inode_list;
439 bool need_writecp = false;
441 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
442 sizeof(struct fsync_inode_entry), NULL);
443 if (!fsync_entry_slab)
446 INIT_LIST_HEAD(&inode_list);
448 /* step #1: find fsynced inode numbers */
449 sbi->por_doing = true;
450 err = find_fsync_dnodes(sbi, &inode_list);
454 if (list_empty(&inode_list))
459 /* step #2: recover data */
460 err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
461 f2fs_bug_on(!list_empty(&inode_list));
463 destroy_fsync_dnodes(&inode_list);
464 kmem_cache_destroy(fsync_entry_slab);
465 sbi->por_doing = false;
466 if (!err && need_writecp)
467 write_checkpoint(sbi, false);