1/*
2 * fs/f2fs/recovery.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 *             http://www.samsung.com/
6 *
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.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include "f2fs.h"
14#include "node.h"
15#include "segment.h"
16
17/*
18 * Roll forward recovery scenarios.
19 *
20 * [Term] F: fsync_mark, D: dentry_mark
21 *
22 * 1. inode(x) | CP | inode(x) | dnode(F)
23 * -> Update the latest inode(x).
24 *
25 * 2. inode(x) | CP | inode(F) | dnode(F)
26 * -> No problem.
27 *
28 * 3. inode(x) | CP | dnode(F) | inode(x)
29 * -> Recover to the latest dnode(F), and drop the last inode(x)
30 *
31 * 4. inode(x) | CP | dnode(F) | inode(F)
32 * -> No problem.
33 *
34 * 5. CP | inode(x) | dnode(F)
35 * -> The inode(DF) was missing. Should drop this dnode(F).
36 *
37 * 6. CP | inode(DF) | dnode(F)
38 * -> No problem.
39 *
40 * 7. CP | dnode(F) | inode(DF)
41 * -> If f2fs_iget fails, then goto next to find inode(DF).
42 *
43 * 8. CP | dnode(F) | inode(x)
44 * -> If f2fs_iget fails, then goto next to find inode(DF).
45 *    But it will fail due to no inode(DF).
46 */
47
48static struct kmem_cache *fsync_entry_slab;
49
50bool space_for_roll_forward(struct f2fs_sb_info *sbi)
51{
52	if (sbi->last_valid_block_count + sbi->alloc_valid_block_count
53			> sbi->user_block_count)
54		return false;
55	return true;
56}
57
58static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
59								nid_t ino)
60{
61	struct fsync_inode_entry *entry;
62
63	list_for_each_entry(entry, head, list)
64		if (entry->inode->i_ino == ino)
65			return entry;
66
67	return NULL;
68}
69
70static int recover_dentry(struct inode *inode, struct page *ipage)
71{
72	struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
73	nid_t pino = le32_to_cpu(raw_inode->i_pino);
74	struct f2fs_dir_entry *de;
75	struct qstr name;
76	struct page *page;
77	struct inode *dir, *einode;
78	int err = 0;
79
80	dir = f2fs_iget(inode->i_sb, pino);
81	if (IS_ERR(dir)) {
82		err = PTR_ERR(dir);
83		goto out;
84	}
85
86	name.len = le32_to_cpu(raw_inode->i_namelen);
87	name.name = raw_inode->i_name;
88
89	if (unlikely(name.len > F2FS_NAME_LEN)) {
90		WARN_ON(1);
91		err = -ENAMETOOLONG;
92		goto out_err;
93	}
94retry:
95	de = f2fs_find_entry(dir, &name, &page);
96	if (de && inode->i_ino == le32_to_cpu(de->ino))
97		goto out_unmap_put;
98
99	if (de) {
100		einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
101		if (IS_ERR(einode)) {
102			WARN_ON(1);
103			err = PTR_ERR(einode);
104			if (err == -ENOENT)
105				err = -EEXIST;
106			goto out_unmap_put;
107		}
108		err = acquire_orphan_inode(F2FS_I_SB(inode));
109		if (err) {
110			iput(einode);
111			goto out_unmap_put;
112		}
113		f2fs_delete_entry(de, page, dir, einode);
114		iput(einode);
115		goto retry;
116	}
117	err = __f2fs_add_link(dir, &name, inode, inode->i_ino, inode->i_mode);
118	if (err)
119		goto out_err;
120
121	if (is_inode_flag_set(F2FS_I(dir), FI_DELAY_IPUT)) {
122		iput(dir);
123	} else {
124		add_dirty_dir_inode(dir);
125		set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
126	}
127
128	goto out;
129
130out_unmap_put:
131	f2fs_dentry_kunmap(dir, page);
132	f2fs_put_page(page, 0);
133out_err:
134	iput(dir);
135out:
136	f2fs_msg(inode->i_sb, KERN_NOTICE,
137			"%s: ino = %x, name = %s, dir = %lx, err = %d",
138			__func__, ino_of_node(ipage), raw_inode->i_name,
139			IS_ERR(dir) ? 0 : dir->i_ino, err);
140	return err;
141}
142
143static void recover_inode(struct inode *inode, struct page *page)
144{
145	struct f2fs_inode *raw = F2FS_INODE(page);
146
147	inode->i_mode = le16_to_cpu(raw->i_mode);
148	i_size_write(inode, le64_to_cpu(raw->i_size));
149	inode->i_atime.tv_sec = le64_to_cpu(raw->i_mtime);
150	inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
151	inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
152	inode->i_atime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
153	inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
154	inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
155
156	f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
157			ino_of_node(page), F2FS_INODE(page)->i_name);
158}
159
160static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
161{
162	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
163	struct curseg_info *curseg;
164	struct page *page = NULL;
165	block_t blkaddr;
166	int err = 0;
167
168	/* get node pages in the current segment */
169	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
170	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
171
172	ra_meta_pages(sbi, blkaddr, 1, META_POR);
173
174	while (1) {
175		struct fsync_inode_entry *entry;
176
177		if (blkaddr < MAIN_BLKADDR(sbi) || blkaddr >= MAX_BLKADDR(sbi))
178			return 0;
179
180		page = get_meta_page(sbi, blkaddr);
181
182		if (cp_ver != cpver_of_node(page))
183			break;
184
185		if (!is_fsync_dnode(page))
186			goto next;
187
188		entry = get_fsync_inode(head, ino_of_node(page));
189		if (!entry) {
190			if (IS_INODE(page) && is_dent_dnode(page)) {
191				err = recover_inode_page(sbi, page);
192				if (err)
193					break;
194			}
195
196			/* add this fsync inode to the list */
197			entry = kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
198			if (!entry) {
199				err = -ENOMEM;
200				break;
201			}
202			/*
203			 * CP | dnode(F) | inode(DF)
204			 * For this case, we should not give up now.
205			 */
206			entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
207			if (IS_ERR(entry->inode)) {
208				err = PTR_ERR(entry->inode);
209				kmem_cache_free(fsync_entry_slab, entry);
210				if (err == -ENOENT) {
211					err = 0;
212					goto next;
213				}
214				break;
215			}
216			list_add_tail(&entry->list, head);
217		}
218		entry->blkaddr = blkaddr;
219
220		if (IS_INODE(page)) {
221			entry->last_inode = blkaddr;
222			if (is_dent_dnode(page))
223				entry->last_dentry = blkaddr;
224		}
225next:
226		/* check next segment */
227		blkaddr = next_blkaddr_of_node(page);
228		f2fs_put_page(page, 1);
229
230		ra_meta_pages_cond(sbi, blkaddr);
231	}
232	f2fs_put_page(page, 1);
233	return err;
234}
235
236static void destroy_fsync_dnodes(struct list_head *head)
237{
238	struct fsync_inode_entry *entry, *tmp;
239
240	list_for_each_entry_safe(entry, tmp, head, list) {
241		iput(entry->inode);
242		list_del(&entry->list);
243		kmem_cache_free(fsync_entry_slab, entry);
244	}
245}
246
247static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
248			block_t blkaddr, struct dnode_of_data *dn)
249{
250	struct seg_entry *sentry;
251	unsigned int segno = GET_SEGNO(sbi, blkaddr);
252	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
253	struct f2fs_summary_block *sum_node;
254	struct f2fs_summary sum;
255	struct page *sum_page, *node_page;
256	struct dnode_of_data tdn = *dn;
257	nid_t ino, nid;
258	struct inode *inode;
259	unsigned int offset;
260	block_t bidx;
261	int i;
262
263	sentry = get_seg_entry(sbi, segno);
264	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
265		return 0;
266
267	/* Get the previous summary */
268	for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
269		struct curseg_info *curseg = CURSEG_I(sbi, i);
270		if (curseg->segno == segno) {
271			sum = curseg->sum_blk->entries[blkoff];
272			goto got_it;
273		}
274	}
275
276	sum_page = get_sum_page(sbi, segno);
277	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
278	sum = sum_node->entries[blkoff];
279	f2fs_put_page(sum_page, 1);
280got_it:
281	/* Use the locked dnode page and inode */
282	nid = le32_to_cpu(sum.nid);
283	if (dn->inode->i_ino == nid) {
284		tdn.nid = nid;
285		if (!dn->inode_page_locked)
286			lock_page(dn->inode_page);
287		tdn.node_page = dn->inode_page;
288		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
289		goto truncate_out;
290	} else if (dn->nid == nid) {
291		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
292		goto truncate_out;
293	}
294
295	/* Get the node page */
296	node_page = get_node_page(sbi, nid);
297	if (IS_ERR(node_page))
298		return PTR_ERR(node_page);
299
300	offset = ofs_of_node(node_page);
301	ino = ino_of_node(node_page);
302	f2fs_put_page(node_page, 1);
303
304	if (ino != dn->inode->i_ino) {
305		/* Deallocate previous index in the node page */
306		inode = f2fs_iget(sbi->sb, ino);
307		if (IS_ERR(inode))
308			return PTR_ERR(inode);
309	} else {
310		inode = dn->inode;
311	}
312
313	bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
314			le16_to_cpu(sum.ofs_in_node);
315
316	/*
317	 * if inode page is locked, unlock temporarily, but its reference
318	 * count keeps alive.
319	 */
320	if (ino == dn->inode->i_ino && dn->inode_page_locked)
321		unlock_page(dn->inode_page);
322
323	set_new_dnode(&tdn, inode, NULL, NULL, 0);
324	if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
325		goto out;
326
327	if (tdn.data_blkaddr == blkaddr)
328		truncate_data_blocks_range(&tdn, 1);
329
330	f2fs_put_dnode(&tdn);
331out:
332	if (ino != dn->inode->i_ino)
333		iput(inode);
334	else if (dn->inode_page_locked)
335		lock_page(dn->inode_page);
336	return 0;
337
338truncate_out:
339	if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr)
340		truncate_data_blocks_range(&tdn, 1);
341	if (dn->inode->i_ino == nid && !dn->inode_page_locked)
342		unlock_page(dn->inode_page);
343	return 0;
344}
345
346static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
347					struct page *page, block_t blkaddr)
348{
349	struct f2fs_inode_info *fi = F2FS_I(inode);
350	unsigned int start, end;
351	struct dnode_of_data dn;
352	struct f2fs_summary sum;
353	struct node_info ni;
354	int err = 0, recovered = 0;
355
356	/* step 1: recover xattr */
357	if (IS_INODE(page)) {
358		recover_inline_xattr(inode, page);
359	} else if (f2fs_has_xattr_block(ofs_of_node(page))) {
360		/*
361		 * Deprecated; xattr blocks should be found from cold log.
362		 * But, we should remain this for backward compatibility.
363		 */
364		recover_xattr_data(inode, page, blkaddr);
365		goto out;
366	}
367
368	/* step 2: recover inline data */
369	if (recover_inline_data(inode, page))
370		goto out;
371
372	/* step 3: recover data indices */
373	start = start_bidx_of_node(ofs_of_node(page), fi);
374	end = start + ADDRS_PER_PAGE(page, fi);
375
376	f2fs_lock_op(sbi);
377
378	set_new_dnode(&dn, inode, NULL, NULL, 0);
379
380	err = get_dnode_of_data(&dn, start, ALLOC_NODE);
381	if (err) {
382		f2fs_unlock_op(sbi);
383		goto out;
384	}
385
386	f2fs_wait_on_page_writeback(dn.node_page, NODE);
387
388	get_node_info(sbi, dn.nid, &ni);
389	f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
390	f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
391
392	for (; start < end; start++) {
393		block_t src, dest;
394
395		src = datablock_addr(dn.node_page, dn.ofs_in_node);
396		dest = datablock_addr(page, dn.ofs_in_node);
397
398		if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR &&
399			dest >= MAIN_BLKADDR(sbi) && dest < MAX_BLKADDR(sbi)) {
400
401			if (src == NULL_ADDR) {
402				err = reserve_new_block(&dn);
403				/* We should not get -ENOSPC */
404				f2fs_bug_on(sbi, err);
405			}
406
407			/* Check the previous node page having this index */
408			err = check_index_in_prev_nodes(sbi, dest, &dn);
409			if (err)
410				goto err;
411
412			set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
413
414			/* write dummy data page */
415			recover_data_page(sbi, NULL, &sum, src, dest);
416			dn.data_blkaddr = dest;
417			set_data_blkaddr(&dn);
418			f2fs_update_extent_cache(&dn);
419			recovered++;
420		}
421		dn.ofs_in_node++;
422	}
423
424	if (IS_INODE(dn.node_page))
425		sync_inode_page(&dn);
426
427	copy_node_footer(dn.node_page, page);
428	fill_node_footer(dn.node_page, dn.nid, ni.ino,
429					ofs_of_node(page), false);
430	set_page_dirty(dn.node_page);
431err:
432	f2fs_put_dnode(&dn);
433	f2fs_unlock_op(sbi);
434out:
435	f2fs_msg(sbi->sb, KERN_NOTICE,
436		"recover_data: ino = %lx, recovered = %d blocks, err = %d",
437		inode->i_ino, recovered, err);
438	return err;
439}
440
441static int recover_data(struct f2fs_sb_info *sbi,
442				struct list_head *head, int type)
443{
444	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
445	struct curseg_info *curseg;
446	struct page *page = NULL;
447	int err = 0;
448	block_t blkaddr;
449
450	/* get node pages in the current segment */
451	curseg = CURSEG_I(sbi, type);
452	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
453
454	while (1) {
455		struct fsync_inode_entry *entry;
456
457		if (blkaddr < MAIN_BLKADDR(sbi) || blkaddr >= MAX_BLKADDR(sbi))
458			break;
459
460		ra_meta_pages_cond(sbi, blkaddr);
461
462		page = get_meta_page(sbi, blkaddr);
463
464		if (cp_ver != cpver_of_node(page)) {
465			f2fs_put_page(page, 1);
466			break;
467		}
468
469		entry = get_fsync_inode(head, ino_of_node(page));
470		if (!entry)
471			goto next;
472		/*
473		 * inode(x) | CP | inode(x) | dnode(F)
474		 * In this case, we can lose the latest inode(x).
475		 * So, call recover_inode for the inode update.
476		 */
477		if (entry->last_inode == blkaddr)
478			recover_inode(entry->inode, page);
479		if (entry->last_dentry == blkaddr) {
480			err = recover_dentry(entry->inode, page);
481			if (err) {
482				f2fs_put_page(page, 1);
483				break;
484			}
485		}
486		err = do_recover_data(sbi, entry->inode, page, blkaddr);
487		if (err) {
488			f2fs_put_page(page, 1);
489			break;
490		}
491
492		if (entry->blkaddr == blkaddr) {
493			iput(entry->inode);
494			list_del(&entry->list);
495			kmem_cache_free(fsync_entry_slab, entry);
496		}
497next:
498		/* check next segment */
499		blkaddr = next_blkaddr_of_node(page);
500		f2fs_put_page(page, 1);
501	}
502	if (!err)
503		allocate_new_segments(sbi);
504	return err;
505}
506
507int recover_fsync_data(struct f2fs_sb_info *sbi)
508{
509	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
510	struct list_head inode_list;
511	block_t blkaddr;
512	int err;
513	bool need_writecp = false;
514
515	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
516			sizeof(struct fsync_inode_entry));
517	if (!fsync_entry_slab)
518		return -ENOMEM;
519
520	INIT_LIST_HEAD(&inode_list);
521
522	/* step #1: find fsynced inode numbers */
523	set_sbi_flag(sbi, SBI_POR_DOING);
524
525	/* prevent checkpoint */
526	mutex_lock(&sbi->cp_mutex);
527
528	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
529
530	err = find_fsync_dnodes(sbi, &inode_list);
531	if (err)
532		goto out;
533
534	if (list_empty(&inode_list))
535		goto out;
536
537	need_writecp = true;
538
539	/* step #2: recover data */
540	err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
541	if (!err)
542		f2fs_bug_on(sbi, !list_empty(&inode_list));
543out:
544	destroy_fsync_dnodes(&inode_list);
545	kmem_cache_destroy(fsync_entry_slab);
546
547	/* truncate meta pages to be used by the recovery */
548	truncate_inode_pages_range(META_MAPPING(sbi),
549			MAIN_BLKADDR(sbi) << PAGE_CACHE_SHIFT, -1);
550
551	if (err) {
552		truncate_inode_pages_final(NODE_MAPPING(sbi));
553		truncate_inode_pages_final(META_MAPPING(sbi));
554	}
555
556	clear_sbi_flag(sbi, SBI_POR_DOING);
557	if (err) {
558		discard_next_dnode(sbi, blkaddr);
559
560		/* Flush all the NAT/SIT pages */
561		while (get_pages(sbi, F2FS_DIRTY_META))
562			sync_meta_pages(sbi, META, LONG_MAX);
563		set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
564		mutex_unlock(&sbi->cp_mutex);
565	} else if (need_writecp) {
566		struct cp_control cpc = {
567			.reason = CP_RECOVERY,
568		};
569		mutex_unlock(&sbi->cp_mutex);
570		write_checkpoint(sbi, &cpc);
571	} else {
572		mutex_unlock(&sbi->cp_mutex);
573	}
574	return err;
575}
576