1/*
2 * Copyright IBM Corporation, 2007
3 * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 *
13 */
14
15#include <linux/slab.h>
16#include "ext4_jbd2.h"
17#include "ext4_extents.h"
18
19/*
20 * The contiguous blocks details which can be
21 * represented by a single extent
22 */
23struct migrate_struct {
24	ext4_lblk_t first_block, last_block, curr_block;
25	ext4_fsblk_t first_pblock, last_pblock;
26};
27
28static int finish_range(handle_t *handle, struct inode *inode,
29				struct migrate_struct *lb)
30
31{
32	int retval = 0, needed;
33	struct ext4_extent newext;
34	struct ext4_ext_path *path;
35	if (lb->first_pblock == 0)
36		return 0;
37
38	/* Add the extent to temp inode*/
39	newext.ee_block = cpu_to_le32(lb->first_block);
40	newext.ee_len   = cpu_to_le16(lb->last_block - lb->first_block + 1);
41	ext4_ext_store_pblock(&newext, lb->first_pblock);
42	/* Locking only for convinience since we are operating on temp inode */
43	down_write(&EXT4_I(inode)->i_data_sem);
44	path = ext4_find_extent(inode, lb->first_block, NULL, 0);
45	if (IS_ERR(path)) {
46		retval = PTR_ERR(path);
47		path = NULL;
48		goto err_out;
49	}
50
51	/*
52	 * Calculate the credit needed to inserting this extent
53	 * Since we are doing this in loop we may accumalate extra
54	 * credit. But below we try to not accumalate too much
55	 * of them by restarting the journal.
56	 */
57	needed = ext4_ext_calc_credits_for_single_extent(inode,
58		    lb->last_block - lb->first_block + 1, path);
59
60	/*
61	 * Make sure the credit we accumalated is not really high
62	 */
63	if (needed && ext4_handle_has_enough_credits(handle,
64						EXT4_RESERVE_TRANS_BLOCKS)) {
65		up_write((&EXT4_I(inode)->i_data_sem));
66		retval = ext4_journal_restart(handle, needed);
67		down_write((&EXT4_I(inode)->i_data_sem));
68		if (retval)
69			goto err_out;
70	} else if (needed) {
71		retval = ext4_journal_extend(handle, needed);
72		if (retval) {
73			/*
74			 * IF not able to extend the journal restart the journal
75			 */
76			up_write((&EXT4_I(inode)->i_data_sem));
77			retval = ext4_journal_restart(handle, needed);
78			down_write((&EXT4_I(inode)->i_data_sem));
79			if (retval)
80				goto err_out;
81		}
82	}
83	retval = ext4_ext_insert_extent(handle, inode, &path, &newext, 0);
84err_out:
85	up_write((&EXT4_I(inode)->i_data_sem));
86	ext4_ext_drop_refs(path);
87	kfree(path);
88	lb->first_pblock = 0;
89	return retval;
90}
91
92static int update_extent_range(handle_t *handle, struct inode *inode,
93			       ext4_fsblk_t pblock, struct migrate_struct *lb)
94{
95	int retval;
96	/*
97	 * See if we can add on to the existing range (if it exists)
98	 */
99	if (lb->first_pblock &&
100		(lb->last_pblock+1 == pblock) &&
101		(lb->last_block+1 == lb->curr_block)) {
102		lb->last_pblock = pblock;
103		lb->last_block = lb->curr_block;
104		lb->curr_block++;
105		return 0;
106	}
107	/*
108	 * Start a new range.
109	 */
110	retval = finish_range(handle, inode, lb);
111	lb->first_pblock = lb->last_pblock = pblock;
112	lb->first_block = lb->last_block = lb->curr_block;
113	lb->curr_block++;
114	return retval;
115}
116
117static int update_ind_extent_range(handle_t *handle, struct inode *inode,
118				   ext4_fsblk_t pblock,
119				   struct migrate_struct *lb)
120{
121	struct buffer_head *bh;
122	__le32 *i_data;
123	int i, retval = 0;
124	unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
125
126	bh = sb_bread(inode->i_sb, pblock);
127	if (!bh)
128		return -EIO;
129
130	i_data = (__le32 *)bh->b_data;
131	for (i = 0; i < max_entries; i++) {
132		if (i_data[i]) {
133			retval = update_extent_range(handle, inode,
134						le32_to_cpu(i_data[i]), lb);
135			if (retval)
136				break;
137		} else {
138			lb->curr_block++;
139		}
140	}
141	put_bh(bh);
142	return retval;
143
144}
145
146static int update_dind_extent_range(handle_t *handle, struct inode *inode,
147				    ext4_fsblk_t pblock,
148				    struct migrate_struct *lb)
149{
150	struct buffer_head *bh;
151	__le32 *i_data;
152	int i, retval = 0;
153	unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
154
155	bh = sb_bread(inode->i_sb, pblock);
156	if (!bh)
157		return -EIO;
158
159	i_data = (__le32 *)bh->b_data;
160	for (i = 0; i < max_entries; i++) {
161		if (i_data[i]) {
162			retval = update_ind_extent_range(handle, inode,
163						le32_to_cpu(i_data[i]), lb);
164			if (retval)
165				break;
166		} else {
167			/* Only update the file block number */
168			lb->curr_block += max_entries;
169		}
170	}
171	put_bh(bh);
172	return retval;
173
174}
175
176static int update_tind_extent_range(handle_t *handle, struct inode *inode,
177				    ext4_fsblk_t pblock,
178				    struct migrate_struct *lb)
179{
180	struct buffer_head *bh;
181	__le32 *i_data;
182	int i, retval = 0;
183	unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
184
185	bh = sb_bread(inode->i_sb, pblock);
186	if (!bh)
187		return -EIO;
188
189	i_data = (__le32 *)bh->b_data;
190	for (i = 0; i < max_entries; i++) {
191		if (i_data[i]) {
192			retval = update_dind_extent_range(handle, inode,
193						le32_to_cpu(i_data[i]), lb);
194			if (retval)
195				break;
196		} else {
197			/* Only update the file block number */
198			lb->curr_block += max_entries * max_entries;
199		}
200	}
201	put_bh(bh);
202	return retval;
203
204}
205
206static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
207{
208	int retval = 0, needed;
209
210	if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
211		return 0;
212	/*
213	 * We are freeing a blocks. During this we touch
214	 * superblock, group descriptor and block bitmap.
215	 * So allocate a credit of 3. We may update
216	 * quota (user and group).
217	 */
218	needed = 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
219
220	if (ext4_journal_extend(handle, needed) != 0)
221		retval = ext4_journal_restart(handle, needed);
222
223	return retval;
224}
225
226static int free_dind_blocks(handle_t *handle,
227				struct inode *inode, __le32 i_data)
228{
229	int i;
230	__le32 *tmp_idata;
231	struct buffer_head *bh;
232	unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
233
234	bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
235	if (!bh)
236		return -EIO;
237
238	tmp_idata = (__le32 *)bh->b_data;
239	for (i = 0; i < max_entries; i++) {
240		if (tmp_idata[i]) {
241			extend_credit_for_blkdel(handle, inode);
242			ext4_free_blocks(handle, inode, NULL,
243					 le32_to_cpu(tmp_idata[i]), 1,
244					 EXT4_FREE_BLOCKS_METADATA |
245					 EXT4_FREE_BLOCKS_FORGET);
246		}
247	}
248	put_bh(bh);
249	extend_credit_for_blkdel(handle, inode);
250	ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
251			 EXT4_FREE_BLOCKS_METADATA |
252			 EXT4_FREE_BLOCKS_FORGET);
253	return 0;
254}
255
256static int free_tind_blocks(handle_t *handle,
257				struct inode *inode, __le32 i_data)
258{
259	int i, retval = 0;
260	__le32 *tmp_idata;
261	struct buffer_head *bh;
262	unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
263
264	bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
265	if (!bh)
266		return -EIO;
267
268	tmp_idata = (__le32 *)bh->b_data;
269	for (i = 0; i < max_entries; i++) {
270		if (tmp_idata[i]) {
271			retval = free_dind_blocks(handle,
272					inode, tmp_idata[i]);
273			if (retval) {
274				put_bh(bh);
275				return retval;
276			}
277		}
278	}
279	put_bh(bh);
280	extend_credit_for_blkdel(handle, inode);
281	ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
282			 EXT4_FREE_BLOCKS_METADATA |
283			 EXT4_FREE_BLOCKS_FORGET);
284	return 0;
285}
286
287static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
288{
289	int retval;
290
291	/* ei->i_data[EXT4_IND_BLOCK] */
292	if (i_data[0]) {
293		extend_credit_for_blkdel(handle, inode);
294		ext4_free_blocks(handle, inode, NULL,
295				le32_to_cpu(i_data[0]), 1,
296				 EXT4_FREE_BLOCKS_METADATA |
297				 EXT4_FREE_BLOCKS_FORGET);
298	}
299
300	/* ei->i_data[EXT4_DIND_BLOCK] */
301	if (i_data[1]) {
302		retval = free_dind_blocks(handle, inode, i_data[1]);
303		if (retval)
304			return retval;
305	}
306
307	/* ei->i_data[EXT4_TIND_BLOCK] */
308	if (i_data[2]) {
309		retval = free_tind_blocks(handle, inode, i_data[2]);
310		if (retval)
311			return retval;
312	}
313	return 0;
314}
315
316static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
317						struct inode *tmp_inode)
318{
319	int retval;
320	__le32	i_data[3];
321	struct ext4_inode_info *ei = EXT4_I(inode);
322	struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
323
324	/*
325	 * One credit accounted for writing the
326	 * i_data field of the original inode
327	 */
328	retval = ext4_journal_extend(handle, 1);
329	if (retval) {
330		retval = ext4_journal_restart(handle, 1);
331		if (retval)
332			goto err_out;
333	}
334
335	i_data[0] = ei->i_data[EXT4_IND_BLOCK];
336	i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
337	i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
338
339	down_write(&EXT4_I(inode)->i_data_sem);
340	/*
341	 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
342	 * happened after we started the migrate. We need to
343	 * fail the migrate
344	 */
345	if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) {
346		retval = -EAGAIN;
347		up_write(&EXT4_I(inode)->i_data_sem);
348		goto err_out;
349	} else
350		ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
351	/*
352	 * We have the extent map build with the tmp inode.
353	 * Now copy the i_data across
354	 */
355	ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
356	memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
357
358	/*
359	 * Update i_blocks with the new blocks that got
360	 * allocated while adding extents for extent index
361	 * blocks.
362	 *
363	 * While converting to extents we need not
364	 * update the orignal inode i_blocks for extent blocks
365	 * via quota APIs. The quota update happened via tmp_inode already.
366	 */
367	spin_lock(&inode->i_lock);
368	inode->i_blocks += tmp_inode->i_blocks;
369	spin_unlock(&inode->i_lock);
370	up_write(&EXT4_I(inode)->i_data_sem);
371
372	/*
373	 * We mark the inode dirty after, because we decrement the
374	 * i_blocks when freeing the indirect meta-data blocks
375	 */
376	retval = free_ind_block(handle, inode, i_data);
377	ext4_mark_inode_dirty(handle, inode);
378
379err_out:
380	return retval;
381}
382
383static int free_ext_idx(handle_t *handle, struct inode *inode,
384					struct ext4_extent_idx *ix)
385{
386	int i, retval = 0;
387	ext4_fsblk_t block;
388	struct buffer_head *bh;
389	struct ext4_extent_header *eh;
390
391	block = ext4_idx_pblock(ix);
392	bh = sb_bread(inode->i_sb, block);
393	if (!bh)
394		return -EIO;
395
396	eh = (struct ext4_extent_header *)bh->b_data;
397	if (eh->eh_depth != 0) {
398		ix = EXT_FIRST_INDEX(eh);
399		for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
400			retval = free_ext_idx(handle, inode, ix);
401			if (retval)
402				break;
403		}
404	}
405	put_bh(bh);
406	extend_credit_for_blkdel(handle, inode);
407	ext4_free_blocks(handle, inode, NULL, block, 1,
408			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
409	return retval;
410}
411
412/*
413 * Free the extent meta data blocks only
414 */
415static int free_ext_block(handle_t *handle, struct inode *inode)
416{
417	int i, retval = 0;
418	struct ext4_inode_info *ei = EXT4_I(inode);
419	struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
420	struct ext4_extent_idx *ix;
421	if (eh->eh_depth == 0)
422		/*
423		 * No extra blocks allocated for extent meta data
424		 */
425		return 0;
426	ix = EXT_FIRST_INDEX(eh);
427	for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
428		retval = free_ext_idx(handle, inode, ix);
429		if (retval)
430			return retval;
431	}
432	return retval;
433}
434
435int ext4_ext_migrate(struct inode *inode)
436{
437	handle_t *handle;
438	int retval = 0, i;
439	__le32 *i_data;
440	struct ext4_inode_info *ei;
441	struct inode *tmp_inode = NULL;
442	struct migrate_struct lb;
443	unsigned long max_entries;
444	__u32 goal;
445	uid_t owner[2];
446
447	/*
448	 * If the filesystem does not support extents, or the inode
449	 * already is extent-based, error out.
450	 */
451	if (!ext4_has_feature_extents(inode->i_sb) ||
452	    (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
453		return -EINVAL;
454
455	if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
456		/*
457		 * don't migrate fast symlink
458		 */
459		return retval;
460
461	/*
462	 * Worst case we can touch the allocation bitmaps, a bgd
463	 * block, and a block to link in the orphan list.  We do need
464	 * need to worry about credits for modifying the quota inode.
465	 */
466	handle = ext4_journal_start(inode, EXT4_HT_MIGRATE,
467		4 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
468
469	if (IS_ERR(handle)) {
470		retval = PTR_ERR(handle);
471		return retval;
472	}
473	goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *
474		EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;
475	owner[0] = i_uid_read(inode);
476	owner[1] = i_gid_read(inode);
477	tmp_inode = ext4_new_inode(handle, d_inode(inode->i_sb->s_root),
478				   S_IFREG, NULL, goal, owner);
479	if (IS_ERR(tmp_inode)) {
480		retval = PTR_ERR(tmp_inode);
481		ext4_journal_stop(handle);
482		return retval;
483	}
484	i_size_write(tmp_inode, i_size_read(inode));
485	/*
486	 * Set the i_nlink to zero so it will be deleted later
487	 * when we drop inode reference.
488	 */
489	clear_nlink(tmp_inode);
490
491	ext4_ext_tree_init(handle, tmp_inode);
492	ext4_orphan_add(handle, tmp_inode);
493	ext4_journal_stop(handle);
494
495	/*
496	 * start with one credit accounted for
497	 * superblock modification.
498	 *
499	 * For the tmp_inode we already have committed the
500	 * transaction that created the inode. Later as and
501	 * when we add extents we extent the journal
502	 */
503	/*
504	 * Even though we take i_mutex we can still cause block
505	 * allocation via mmap write to holes. If we have allocated
506	 * new blocks we fail migrate.  New block allocation will
507	 * clear EXT4_STATE_EXT_MIGRATE flag.  The flag is updated
508	 * with i_data_sem held to prevent racing with block
509	 * allocation.
510	 */
511	down_read(&EXT4_I(inode)->i_data_sem);
512	ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
513	up_read((&EXT4_I(inode)->i_data_sem));
514
515	handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
516	if (IS_ERR(handle)) {
517		/*
518		 * It is impossible to update on-disk structures without
519		 * a handle, so just rollback in-core changes and live other
520		 * work to orphan_list_cleanup()
521		 */
522		ext4_orphan_del(NULL, tmp_inode);
523		retval = PTR_ERR(handle);
524		goto out;
525	}
526
527	ei = EXT4_I(inode);
528	i_data = ei->i_data;
529	memset(&lb, 0, sizeof(lb));
530
531	/* 32 bit block address 4 bytes */
532	max_entries = inode->i_sb->s_blocksize >> 2;
533	for (i = 0; i < EXT4_NDIR_BLOCKS; i++) {
534		if (i_data[i]) {
535			retval = update_extent_range(handle, tmp_inode,
536						le32_to_cpu(i_data[i]), &lb);
537			if (retval)
538				goto err_out;
539		} else
540			lb.curr_block++;
541	}
542	if (i_data[EXT4_IND_BLOCK]) {
543		retval = update_ind_extent_range(handle, tmp_inode,
544				le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb);
545			if (retval)
546				goto err_out;
547	} else
548		lb.curr_block += max_entries;
549	if (i_data[EXT4_DIND_BLOCK]) {
550		retval = update_dind_extent_range(handle, tmp_inode,
551				le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb);
552			if (retval)
553				goto err_out;
554	} else
555		lb.curr_block += max_entries * max_entries;
556	if (i_data[EXT4_TIND_BLOCK]) {
557		retval = update_tind_extent_range(handle, tmp_inode,
558				le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb);
559			if (retval)
560				goto err_out;
561	}
562	/*
563	 * Build the last extent
564	 */
565	retval = finish_range(handle, tmp_inode, &lb);
566err_out:
567	if (retval)
568		/*
569		 * Failure case delete the extent information with the
570		 * tmp_inode
571		 */
572		free_ext_block(handle, tmp_inode);
573	else {
574		retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
575		if (retval)
576			/*
577			 * if we fail to swap inode data free the extent
578			 * details of the tmp inode
579			 */
580			free_ext_block(handle, tmp_inode);
581	}
582
583	/* We mark the tmp_inode dirty via ext4_ext_tree_init. */
584	if (ext4_journal_extend(handle, 1) != 0)
585		ext4_journal_restart(handle, 1);
586
587	/*
588	 * Mark the tmp_inode as of size zero
589	 */
590	i_size_write(tmp_inode, 0);
591
592	/*
593	 * set the  i_blocks count to zero
594	 * so that the ext4_evict_inode() does the
595	 * right job
596	 *
597	 * We don't need to take the i_lock because
598	 * the inode is not visible to user space.
599	 */
600	tmp_inode->i_blocks = 0;
601
602	/* Reset the extent details */
603	ext4_ext_tree_init(handle, tmp_inode);
604	ext4_journal_stop(handle);
605out:
606	unlock_new_inode(tmp_inode);
607	iput(tmp_inode);
608
609	return retval;
610}
611
612/*
613 * Migrate a simple extent-based inode to use the i_blocks[] array
614 */
615int ext4_ind_migrate(struct inode *inode)
616{
617	struct ext4_extent_header	*eh;
618	struct ext4_super_block		*es = EXT4_SB(inode->i_sb)->s_es;
619	struct ext4_inode_info		*ei = EXT4_I(inode);
620	struct ext4_extent		*ex;
621	unsigned int			i, len;
622	ext4_lblk_t			start, end;
623	ext4_fsblk_t			blk;
624	handle_t			*handle;
625	int				ret;
626
627	if (!ext4_has_feature_extents(inode->i_sb) ||
628	    (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
629		return -EINVAL;
630
631	if (ext4_has_feature_bigalloc(inode->i_sb))
632		return -EOPNOTSUPP;
633
634	/*
635	 * In order to get correct extent info, force all delayed allocation
636	 * blocks to be allocated, otherwise delayed allocation blocks may not
637	 * be reflected and bypass the checks on extent header.
638	 */
639	if (test_opt(inode->i_sb, DELALLOC))
640		ext4_alloc_da_blocks(inode);
641
642	handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
643	if (IS_ERR(handle))
644		return PTR_ERR(handle);
645
646	down_write(&EXT4_I(inode)->i_data_sem);
647	ret = ext4_ext_check_inode(inode);
648	if (ret)
649		goto errout;
650
651	eh = ext_inode_hdr(inode);
652	ex  = EXT_FIRST_EXTENT(eh);
653	if (ext4_blocks_count(es) > EXT4_MAX_BLOCK_FILE_PHYS ||
654	    eh->eh_depth != 0 || le16_to_cpu(eh->eh_entries) > 1) {
655		ret = -EOPNOTSUPP;
656		goto errout;
657	}
658	if (eh->eh_entries == 0)
659		blk = len = start = end = 0;
660	else {
661		len = le16_to_cpu(ex->ee_len);
662		blk = ext4_ext_pblock(ex);
663		start = le32_to_cpu(ex->ee_block);
664		end = start + len - 1;
665		if (end >= EXT4_NDIR_BLOCKS) {
666			ret = -EOPNOTSUPP;
667			goto errout;
668		}
669	}
670
671	ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
672	memset(ei->i_data, 0, sizeof(ei->i_data));
673	for (i = start; i <= end; i++)
674		ei->i_data[i] = cpu_to_le32(blk++);
675	ext4_mark_inode_dirty(handle, inode);
676errout:
677	ext4_journal_stop(handle);
678	up_write(&EXT4_I(inode)->i_data_sem);
679	return ret;
680}
681