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 fsync_inode_entry *entry;
32 list_for_each_entry(entry, head, list)
33 if (entry->inode->i_ino == ino)
39 static int recover_dentry(struct page *ipage, struct inode *inode)
41 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
42 nid_t pino = le32_to_cpu(raw_inode->i_pino);
43 struct f2fs_dir_entry *de;
46 struct inode *dir, *einode;
49 dir = f2fs_iget(inode->i_sb, pino);
55 name.len = le32_to_cpu(raw_inode->i_namelen);
56 name.name = raw_inode->i_name;
58 if (unlikely(name.len > F2FS_NAME_LEN)) {
64 de = f2fs_find_entry(dir, &name, &page);
65 if (de && inode->i_ino == le32_to_cpu(de->ino))
68 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
71 err = PTR_ERR(einode);
76 err = acquire_orphan_inode(F2FS_SB(inode->i_sb));
81 f2fs_delete_entry(de, page, einode);
85 err = __f2fs_add_link(dir, &name, inode);
89 if (is_inode_flag_set(F2FS_I(dir), FI_DELAY_IPUT)) {
92 add_dirty_dir_inode(dir);
93 set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
100 f2fs_put_page(page, 0);
104 f2fs_msg(inode->i_sb, KERN_NOTICE,
105 "%s: ino = %x, name = %s, dir = %lx, err = %d",
106 __func__, ino_of_node(ipage), raw_inode->i_name,
107 IS_ERR(dir) ? 0 : dir->i_ino, err);
111 static int recover_inode(struct inode *inode, struct page *node_page)
113 struct f2fs_inode *raw_inode = F2FS_INODE(node_page);
115 if (!IS_INODE(node_page))
118 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
119 i_size_write(inode, le64_to_cpu(raw_inode->i_size));
120 inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
121 inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
122 inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
123 inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
124 inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
125 inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
127 if (is_dent_dnode(node_page))
128 return recover_dentry(node_page, inode);
130 f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
131 ino_of_node(node_page), raw_inode->i_name);
135 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
137 unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
138 struct curseg_info *curseg;
143 /* get node pages in the current segment */
144 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
145 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
148 page = alloc_page(GFP_F2FS_ZERO);
154 struct fsync_inode_entry *entry;
156 err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC);
162 if (cp_ver != cpver_of_node(page))
165 if (!is_fsync_dnode(page))
168 entry = get_fsync_inode(head, ino_of_node(page));
170 if (IS_INODE(page) && is_dent_dnode(page))
171 set_inode_flag(F2FS_I(entry->inode),
174 if (IS_INODE(page) && is_dent_dnode(page)) {
175 err = recover_inode_page(sbi, page);
180 /* add this fsync inode to the list */
181 entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
187 entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
188 if (IS_ERR(entry->inode)) {
189 err = PTR_ERR(entry->inode);
190 kmem_cache_free(fsync_entry_slab, entry);
193 list_add_tail(&entry->list, head);
195 entry->blkaddr = blkaddr;
197 err = recover_inode(entry->inode, page);
198 if (err && err != -ENOENT)
201 /* check next segment */
202 blkaddr = next_blkaddr_of_node(page);
206 __free_pages(page, 0);
211 static void destroy_fsync_dnodes(struct list_head *head)
213 struct fsync_inode_entry *entry, *tmp;
215 list_for_each_entry_safe(entry, tmp, head, list) {
217 list_del(&entry->list);
218 kmem_cache_free(fsync_entry_slab, entry);
222 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
223 block_t blkaddr, struct dnode_of_data *dn)
225 struct seg_entry *sentry;
226 unsigned int segno = GET_SEGNO(sbi, blkaddr);
227 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
228 struct f2fs_summary_block *sum_node;
229 struct f2fs_summary sum;
230 struct page *sum_page, *node_page;
237 sentry = get_seg_entry(sbi, segno);
238 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
241 /* Get the previous summary */
242 for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
243 struct curseg_info *curseg = CURSEG_I(sbi, i);
244 if (curseg->segno == segno) {
245 sum = curseg->sum_blk->entries[blkoff];
250 sum_page = get_sum_page(sbi, segno);
251 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
252 sum = sum_node->entries[blkoff];
253 f2fs_put_page(sum_page, 1);
255 /* Use the locked dnode page and inode */
256 nid = le32_to_cpu(sum.nid);
257 if (dn->inode->i_ino == nid) {
258 struct dnode_of_data tdn = *dn;
260 tdn.node_page = dn->inode_page;
261 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
262 truncate_data_blocks_range(&tdn, 1);
264 } else if (dn->nid == nid) {
265 struct dnode_of_data tdn = *dn;
266 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
267 truncate_data_blocks_range(&tdn, 1);
271 /* Get the node page */
272 node_page = get_node_page(sbi, nid);
273 if (IS_ERR(node_page))
274 return PTR_ERR(node_page);
276 offset = ofs_of_node(node_page);
277 ino = ino_of_node(node_page);
278 f2fs_put_page(node_page, 1);
280 /* Deallocate previous index in the node page */
281 inode = f2fs_iget(sbi->sb, ino);
283 return PTR_ERR(inode);
285 bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
286 le16_to_cpu(sum.ofs_in_node);
288 truncate_hole(inode, bidx, bidx + 1);
293 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
294 struct page *page, block_t blkaddr)
296 struct f2fs_inode_info *fi = F2FS_I(inode);
297 unsigned int start, end;
298 struct dnode_of_data dn;
299 struct f2fs_summary sum;
301 int err = 0, recovered = 0;
303 recover_inline_xattr(inode, page);
305 if (recover_inline_data(inode, page))
308 if (recover_xattr_data(inode, page, blkaddr))
311 start = start_bidx_of_node(ofs_of_node(page), fi);
312 end = start + ADDRS_PER_PAGE(page, fi);
316 set_new_dnode(&dn, inode, NULL, NULL, 0);
318 err = get_dnode_of_data(&dn, start, ALLOC_NODE);
324 f2fs_wait_on_page_writeback(dn.node_page, NODE);
326 get_node_info(sbi, dn.nid, &ni);
327 f2fs_bug_on(ni.ino != ino_of_node(page));
328 f2fs_bug_on(ofs_of_node(dn.node_page) != ofs_of_node(page));
330 for (; start < end; start++) {
333 src = datablock_addr(dn.node_page, dn.ofs_in_node);
334 dest = datablock_addr(page, dn.ofs_in_node);
336 if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
337 if (src == NULL_ADDR) {
338 err = reserve_new_block(&dn);
339 /* We should not get -ENOSPC */
343 /* Check the previous node page having this index */
344 err = check_index_in_prev_nodes(sbi, dest, &dn);
348 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
350 /* write dummy data page */
351 recover_data_page(sbi, NULL, &sum, src, dest);
352 update_extent_cache(dest, &dn);
358 /* write node page in place */
359 set_summary(&sum, dn.nid, 0, 0);
360 if (IS_INODE(dn.node_page))
361 sync_inode_page(&dn);
363 copy_node_footer(dn.node_page, page);
364 fill_node_footer(dn.node_page, dn.nid, ni.ino,
365 ofs_of_node(page), false);
366 set_page_dirty(dn.node_page);
368 recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr);
373 f2fs_msg(sbi->sb, KERN_NOTICE,
374 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
375 inode->i_ino, recovered, err);
379 static int recover_data(struct f2fs_sb_info *sbi,
380 struct list_head *head, int type)
382 unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
383 struct curseg_info *curseg;
388 /* get node pages in the current segment */
389 curseg = CURSEG_I(sbi, type);
390 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
393 page = alloc_page(GFP_F2FS_ZERO);
400 struct fsync_inode_entry *entry;
402 err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC);
408 if (cp_ver != cpver_of_node(page))
411 entry = get_fsync_inode(head, ino_of_node(page));
415 err = do_recover_data(sbi, entry->inode, page, blkaddr);
419 if (entry->blkaddr == blkaddr) {
421 list_del(&entry->list);
422 kmem_cache_free(fsync_entry_slab, entry);
425 /* check next segment */
426 blkaddr = next_blkaddr_of_node(page);
430 __free_pages(page, 0);
433 allocate_new_segments(sbi);
437 int recover_fsync_data(struct f2fs_sb_info *sbi)
439 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
440 struct list_head inode_list;
443 bool need_writecp = false;
445 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
446 sizeof(struct fsync_inode_entry));
447 if (!fsync_entry_slab)
450 INIT_LIST_HEAD(&inode_list);
452 /* step #1: find fsynced inode numbers */
453 sbi->por_doing = true;
455 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
457 err = find_fsync_dnodes(sbi, &inode_list);
461 if (list_empty(&inode_list))
466 /* step #2: recover data */
467 err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
468 f2fs_bug_on(!list_empty(&inode_list));
470 destroy_fsync_dnodes(&inode_list);
471 kmem_cache_destroy(fsync_entry_slab);
474 truncate_inode_pages_final(NODE_MAPPING(sbi));
475 truncate_inode_pages_final(META_MAPPING(sbi));
478 sbi->por_doing = false;
480 discard_next_dnode(sbi, blkaddr);
482 /* Flush all the NAT/SIT pages */
483 while (get_pages(sbi, F2FS_DIRTY_META))
484 sync_meta_pages(sbi, META, LONG_MAX);
485 } else if (need_writecp) {
486 write_checkpoint(sbi, false);