Loading...
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 original 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}
1// SPDX-License-Identifier: LGPL-2.1
2/*
3 * Copyright IBM Corporation, 2007
4 * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
5 *
6 */
7
8#include <linux/slab.h>
9#include "ext4_jbd2.h"
10#include "ext4_extents.h"
11
12/*
13 * The contiguous blocks details which can be
14 * represented by a single extent
15 */
16struct migrate_struct {
17 ext4_lblk_t first_block, last_block, curr_block;
18 ext4_fsblk_t first_pblock, last_pblock;
19};
20
21static int finish_range(handle_t *handle, struct inode *inode,
22 struct migrate_struct *lb)
23
24{
25 int retval = 0, needed;
26 struct ext4_extent newext;
27 struct ext4_ext_path *path;
28 if (lb->first_pblock == 0)
29 return 0;
30
31 /* Add the extent to temp inode*/
32 newext.ee_block = cpu_to_le32(lb->first_block);
33 newext.ee_len = cpu_to_le16(lb->last_block - lb->first_block + 1);
34 ext4_ext_store_pblock(&newext, lb->first_pblock);
35 /* Locking only for convinience since we are operating on temp inode */
36 down_write(&EXT4_I(inode)->i_data_sem);
37 path = ext4_find_extent(inode, lb->first_block, NULL, 0);
38 if (IS_ERR(path)) {
39 retval = PTR_ERR(path);
40 path = NULL;
41 goto err_out;
42 }
43
44 /*
45 * Calculate the credit needed to inserting this extent
46 * Since we are doing this in loop we may accumalate extra
47 * credit. But below we try to not accumalate too much
48 * of them by restarting the journal.
49 */
50 needed = ext4_ext_calc_credits_for_single_extent(inode,
51 lb->last_block - lb->first_block + 1, path);
52
53 /*
54 * Make sure the credit we accumalated is not really high
55 */
56 if (needed && ext4_handle_has_enough_credits(handle,
57 EXT4_RESERVE_TRANS_BLOCKS)) {
58 up_write((&EXT4_I(inode)->i_data_sem));
59 retval = ext4_journal_restart(handle, needed);
60 down_write((&EXT4_I(inode)->i_data_sem));
61 if (retval)
62 goto err_out;
63 } else if (needed) {
64 retval = ext4_journal_extend(handle, needed);
65 if (retval) {
66 /*
67 * IF not able to extend the journal restart the journal
68 */
69 up_write((&EXT4_I(inode)->i_data_sem));
70 retval = ext4_journal_restart(handle, needed);
71 down_write((&EXT4_I(inode)->i_data_sem));
72 if (retval)
73 goto err_out;
74 }
75 }
76 retval = ext4_ext_insert_extent(handle, inode, &path, &newext, 0);
77err_out:
78 up_write((&EXT4_I(inode)->i_data_sem));
79 ext4_ext_drop_refs(path);
80 kfree(path);
81 lb->first_pblock = 0;
82 return retval;
83}
84
85static int update_extent_range(handle_t *handle, struct inode *inode,
86 ext4_fsblk_t pblock, struct migrate_struct *lb)
87{
88 int retval;
89 /*
90 * See if we can add on to the existing range (if it exists)
91 */
92 if (lb->first_pblock &&
93 (lb->last_pblock+1 == pblock) &&
94 (lb->last_block+1 == lb->curr_block)) {
95 lb->last_pblock = pblock;
96 lb->last_block = lb->curr_block;
97 lb->curr_block++;
98 return 0;
99 }
100 /*
101 * Start a new range.
102 */
103 retval = finish_range(handle, inode, lb);
104 lb->first_pblock = lb->last_pblock = pblock;
105 lb->first_block = lb->last_block = lb->curr_block;
106 lb->curr_block++;
107 return retval;
108}
109
110static int update_ind_extent_range(handle_t *handle, struct inode *inode,
111 ext4_fsblk_t pblock,
112 struct migrate_struct *lb)
113{
114 struct buffer_head *bh;
115 __le32 *i_data;
116 int i, retval = 0;
117 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
118
119 bh = ext4_sb_bread(inode->i_sb, pblock, 0);
120 if (IS_ERR(bh))
121 return PTR_ERR(bh);
122
123 i_data = (__le32 *)bh->b_data;
124 for (i = 0; i < max_entries; i++) {
125 if (i_data[i]) {
126 retval = update_extent_range(handle, inode,
127 le32_to_cpu(i_data[i]), lb);
128 if (retval)
129 break;
130 } else {
131 lb->curr_block++;
132 }
133 }
134 put_bh(bh);
135 return retval;
136
137}
138
139static int update_dind_extent_range(handle_t *handle, struct inode *inode,
140 ext4_fsblk_t pblock,
141 struct migrate_struct *lb)
142{
143 struct buffer_head *bh;
144 __le32 *i_data;
145 int i, retval = 0;
146 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
147
148 bh = ext4_sb_bread(inode->i_sb, pblock, 0);
149 if (IS_ERR(bh))
150 return PTR_ERR(bh);
151
152 i_data = (__le32 *)bh->b_data;
153 for (i = 0; i < max_entries; i++) {
154 if (i_data[i]) {
155 retval = update_ind_extent_range(handle, inode,
156 le32_to_cpu(i_data[i]), lb);
157 if (retval)
158 break;
159 } else {
160 /* Only update the file block number */
161 lb->curr_block += max_entries;
162 }
163 }
164 put_bh(bh);
165 return retval;
166
167}
168
169static int update_tind_extent_range(handle_t *handle, struct inode *inode,
170 ext4_fsblk_t pblock,
171 struct migrate_struct *lb)
172{
173 struct buffer_head *bh;
174 __le32 *i_data;
175 int i, retval = 0;
176 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
177
178 bh = ext4_sb_bread(inode->i_sb, pblock, 0);
179 if (IS_ERR(bh))
180 return PTR_ERR(bh);
181
182 i_data = (__le32 *)bh->b_data;
183 for (i = 0; i < max_entries; i++) {
184 if (i_data[i]) {
185 retval = update_dind_extent_range(handle, inode,
186 le32_to_cpu(i_data[i]), lb);
187 if (retval)
188 break;
189 } else {
190 /* Only update the file block number */
191 lb->curr_block += max_entries * max_entries;
192 }
193 }
194 put_bh(bh);
195 return retval;
196
197}
198
199static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
200{
201 int retval = 0, needed;
202
203 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
204 return 0;
205 /*
206 * We are freeing a blocks. During this we touch
207 * superblock, group descriptor and block bitmap.
208 * So allocate a credit of 3. We may update
209 * quota (user and group).
210 */
211 needed = 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
212
213 if (ext4_journal_extend(handle, needed) != 0)
214 retval = ext4_journal_restart(handle, needed);
215
216 return retval;
217}
218
219static int free_dind_blocks(handle_t *handle,
220 struct inode *inode, __le32 i_data)
221{
222 int i;
223 __le32 *tmp_idata;
224 struct buffer_head *bh;
225 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
226
227 bh = ext4_sb_bread(inode->i_sb, le32_to_cpu(i_data), 0);
228 if (IS_ERR(bh))
229 return PTR_ERR(bh);
230
231 tmp_idata = (__le32 *)bh->b_data;
232 for (i = 0; i < max_entries; i++) {
233 if (tmp_idata[i]) {
234 extend_credit_for_blkdel(handle, inode);
235 ext4_free_blocks(handle, inode, NULL,
236 le32_to_cpu(tmp_idata[i]), 1,
237 EXT4_FREE_BLOCKS_METADATA |
238 EXT4_FREE_BLOCKS_FORGET);
239 }
240 }
241 put_bh(bh);
242 extend_credit_for_blkdel(handle, inode);
243 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
244 EXT4_FREE_BLOCKS_METADATA |
245 EXT4_FREE_BLOCKS_FORGET);
246 return 0;
247}
248
249static int free_tind_blocks(handle_t *handle,
250 struct inode *inode, __le32 i_data)
251{
252 int i, retval = 0;
253 __le32 *tmp_idata;
254 struct buffer_head *bh;
255 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
256
257 bh = ext4_sb_bread(inode->i_sb, le32_to_cpu(i_data), 0);
258 if (IS_ERR(bh))
259 return PTR_ERR(bh);
260
261 tmp_idata = (__le32 *)bh->b_data;
262 for (i = 0; i < max_entries; i++) {
263 if (tmp_idata[i]) {
264 retval = free_dind_blocks(handle,
265 inode, tmp_idata[i]);
266 if (retval) {
267 put_bh(bh);
268 return retval;
269 }
270 }
271 }
272 put_bh(bh);
273 extend_credit_for_blkdel(handle, inode);
274 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
275 EXT4_FREE_BLOCKS_METADATA |
276 EXT4_FREE_BLOCKS_FORGET);
277 return 0;
278}
279
280static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
281{
282 int retval;
283
284 /* ei->i_data[EXT4_IND_BLOCK] */
285 if (i_data[0]) {
286 extend_credit_for_blkdel(handle, inode);
287 ext4_free_blocks(handle, inode, NULL,
288 le32_to_cpu(i_data[0]), 1,
289 EXT4_FREE_BLOCKS_METADATA |
290 EXT4_FREE_BLOCKS_FORGET);
291 }
292
293 /* ei->i_data[EXT4_DIND_BLOCK] */
294 if (i_data[1]) {
295 retval = free_dind_blocks(handle, inode, i_data[1]);
296 if (retval)
297 return retval;
298 }
299
300 /* ei->i_data[EXT4_TIND_BLOCK] */
301 if (i_data[2]) {
302 retval = free_tind_blocks(handle, inode, i_data[2]);
303 if (retval)
304 return retval;
305 }
306 return 0;
307}
308
309static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
310 struct inode *tmp_inode)
311{
312 int retval;
313 __le32 i_data[3];
314 struct ext4_inode_info *ei = EXT4_I(inode);
315 struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
316
317 /*
318 * One credit accounted for writing the
319 * i_data field of the original inode
320 */
321 retval = ext4_journal_extend(handle, 1);
322 if (retval) {
323 retval = ext4_journal_restart(handle, 1);
324 if (retval)
325 goto err_out;
326 }
327
328 i_data[0] = ei->i_data[EXT4_IND_BLOCK];
329 i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
330 i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
331
332 down_write(&EXT4_I(inode)->i_data_sem);
333 /*
334 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
335 * happened after we started the migrate. We need to
336 * fail the migrate
337 */
338 if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) {
339 retval = -EAGAIN;
340 up_write(&EXT4_I(inode)->i_data_sem);
341 goto err_out;
342 } else
343 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
344 /*
345 * We have the extent map build with the tmp inode.
346 * Now copy the i_data across
347 */
348 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
349 memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
350
351 /*
352 * Update i_blocks with the new blocks that got
353 * allocated while adding extents for extent index
354 * blocks.
355 *
356 * While converting to extents we need not
357 * update the original inode i_blocks for extent blocks
358 * via quota APIs. The quota update happened via tmp_inode already.
359 */
360 spin_lock(&inode->i_lock);
361 inode->i_blocks += tmp_inode->i_blocks;
362 spin_unlock(&inode->i_lock);
363 up_write(&EXT4_I(inode)->i_data_sem);
364
365 /*
366 * We mark the inode dirty after, because we decrement the
367 * i_blocks when freeing the indirect meta-data blocks
368 */
369 retval = free_ind_block(handle, inode, i_data);
370 ext4_mark_inode_dirty(handle, inode);
371
372err_out:
373 return retval;
374}
375
376static int free_ext_idx(handle_t *handle, struct inode *inode,
377 struct ext4_extent_idx *ix)
378{
379 int i, retval = 0;
380 ext4_fsblk_t block;
381 struct buffer_head *bh;
382 struct ext4_extent_header *eh;
383
384 block = ext4_idx_pblock(ix);
385 bh = ext4_sb_bread(inode->i_sb, block, 0);
386 if (IS_ERR(bh))
387 return PTR_ERR(bh);
388
389 eh = (struct ext4_extent_header *)bh->b_data;
390 if (eh->eh_depth != 0) {
391 ix = EXT_FIRST_INDEX(eh);
392 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
393 retval = free_ext_idx(handle, inode, ix);
394 if (retval)
395 break;
396 }
397 }
398 put_bh(bh);
399 extend_credit_for_blkdel(handle, inode);
400 ext4_free_blocks(handle, inode, NULL, block, 1,
401 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
402 return retval;
403}
404
405/*
406 * Free the extent meta data blocks only
407 */
408static int free_ext_block(handle_t *handle, struct inode *inode)
409{
410 int i, retval = 0;
411 struct ext4_inode_info *ei = EXT4_I(inode);
412 struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
413 struct ext4_extent_idx *ix;
414 if (eh->eh_depth == 0)
415 /*
416 * No extra blocks allocated for extent meta data
417 */
418 return 0;
419 ix = EXT_FIRST_INDEX(eh);
420 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
421 retval = free_ext_idx(handle, inode, ix);
422 if (retval)
423 return retval;
424 }
425 return retval;
426}
427
428int ext4_ext_migrate(struct inode *inode)
429{
430 handle_t *handle;
431 int retval = 0, i;
432 __le32 *i_data;
433 struct ext4_inode_info *ei;
434 struct inode *tmp_inode = NULL;
435 struct migrate_struct lb;
436 unsigned long max_entries;
437 __u32 goal;
438 uid_t owner[2];
439
440 /*
441 * If the filesystem does not support extents, or the inode
442 * already is extent-based, error out.
443 */
444 if (!ext4_has_feature_extents(inode->i_sb) ||
445 (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
446 return -EINVAL;
447
448 if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
449 /*
450 * don't migrate fast symlink
451 */
452 return retval;
453
454 /*
455 * Worst case we can touch the allocation bitmaps, a bgd
456 * block, and a block to link in the orphan list. We do need
457 * need to worry about credits for modifying the quota inode.
458 */
459 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE,
460 4 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
461
462 if (IS_ERR(handle)) {
463 retval = PTR_ERR(handle);
464 return retval;
465 }
466 goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *
467 EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;
468 owner[0] = i_uid_read(inode);
469 owner[1] = i_gid_read(inode);
470 tmp_inode = ext4_new_inode(handle, d_inode(inode->i_sb->s_root),
471 S_IFREG, NULL, goal, owner, 0);
472 if (IS_ERR(tmp_inode)) {
473 retval = PTR_ERR(tmp_inode);
474 ext4_journal_stop(handle);
475 return retval;
476 }
477 i_size_write(tmp_inode, i_size_read(inode));
478 /*
479 * Set the i_nlink to zero so it will be deleted later
480 * when we drop inode reference.
481 */
482 clear_nlink(tmp_inode);
483
484 ext4_ext_tree_init(handle, tmp_inode);
485 ext4_orphan_add(handle, tmp_inode);
486 ext4_journal_stop(handle);
487
488 /*
489 * start with one credit accounted for
490 * superblock modification.
491 *
492 * For the tmp_inode we already have committed the
493 * transaction that created the inode. Later as and
494 * when we add extents we extent the journal
495 */
496 /*
497 * Even though we take i_mutex we can still cause block
498 * allocation via mmap write to holes. If we have allocated
499 * new blocks we fail migrate. New block allocation will
500 * clear EXT4_STATE_EXT_MIGRATE flag. The flag is updated
501 * with i_data_sem held to prevent racing with block
502 * allocation.
503 */
504 down_read(&EXT4_I(inode)->i_data_sem);
505 ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
506 up_read((&EXT4_I(inode)->i_data_sem));
507
508 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
509 if (IS_ERR(handle)) {
510 /*
511 * It is impossible to update on-disk structures without
512 * a handle, so just rollback in-core changes and live other
513 * work to orphan_list_cleanup()
514 */
515 ext4_orphan_del(NULL, tmp_inode);
516 retval = PTR_ERR(handle);
517 goto out;
518 }
519
520 ei = EXT4_I(inode);
521 i_data = ei->i_data;
522 memset(&lb, 0, sizeof(lb));
523
524 /* 32 bit block address 4 bytes */
525 max_entries = inode->i_sb->s_blocksize >> 2;
526 for (i = 0; i < EXT4_NDIR_BLOCKS; i++) {
527 if (i_data[i]) {
528 retval = update_extent_range(handle, tmp_inode,
529 le32_to_cpu(i_data[i]), &lb);
530 if (retval)
531 goto err_out;
532 } else
533 lb.curr_block++;
534 }
535 if (i_data[EXT4_IND_BLOCK]) {
536 retval = update_ind_extent_range(handle, tmp_inode,
537 le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb);
538 if (retval)
539 goto err_out;
540 } else
541 lb.curr_block += max_entries;
542 if (i_data[EXT4_DIND_BLOCK]) {
543 retval = update_dind_extent_range(handle, tmp_inode,
544 le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb);
545 if (retval)
546 goto err_out;
547 } else
548 lb.curr_block += max_entries * max_entries;
549 if (i_data[EXT4_TIND_BLOCK]) {
550 retval = update_tind_extent_range(handle, tmp_inode,
551 le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb);
552 if (retval)
553 goto err_out;
554 }
555 /*
556 * Build the last extent
557 */
558 retval = finish_range(handle, tmp_inode, &lb);
559err_out:
560 if (retval)
561 /*
562 * Failure case delete the extent information with the
563 * tmp_inode
564 */
565 free_ext_block(handle, tmp_inode);
566 else {
567 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
568 if (retval)
569 /*
570 * if we fail to swap inode data free the extent
571 * details of the tmp inode
572 */
573 free_ext_block(handle, tmp_inode);
574 }
575
576 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
577 if (ext4_journal_extend(handle, 1) != 0)
578 ext4_journal_restart(handle, 1);
579
580 /*
581 * Mark the tmp_inode as of size zero
582 */
583 i_size_write(tmp_inode, 0);
584
585 /*
586 * set the i_blocks count to zero
587 * so that the ext4_evict_inode() does the
588 * right job
589 *
590 * We don't need to take the i_lock because
591 * the inode is not visible to user space.
592 */
593 tmp_inode->i_blocks = 0;
594
595 /* Reset the extent details */
596 ext4_ext_tree_init(handle, tmp_inode);
597 ext4_journal_stop(handle);
598out:
599 unlock_new_inode(tmp_inode);
600 iput(tmp_inode);
601
602 return retval;
603}
604
605/*
606 * Migrate a simple extent-based inode to use the i_blocks[] array
607 */
608int ext4_ind_migrate(struct inode *inode)
609{
610 struct ext4_extent_header *eh;
611 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
612 struct ext4_inode_info *ei = EXT4_I(inode);
613 struct ext4_extent *ex;
614 unsigned int i, len;
615 ext4_lblk_t start, end;
616 ext4_fsblk_t blk;
617 handle_t *handle;
618 int ret;
619
620 if (!ext4_has_feature_extents(inode->i_sb) ||
621 (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
622 return -EINVAL;
623
624 if (ext4_has_feature_bigalloc(inode->i_sb))
625 return -EOPNOTSUPP;
626
627 /*
628 * In order to get correct extent info, force all delayed allocation
629 * blocks to be allocated, otherwise delayed allocation blocks may not
630 * be reflected and bypass the checks on extent header.
631 */
632 if (test_opt(inode->i_sb, DELALLOC))
633 ext4_alloc_da_blocks(inode);
634
635 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
636 if (IS_ERR(handle))
637 return PTR_ERR(handle);
638
639 down_write(&EXT4_I(inode)->i_data_sem);
640 ret = ext4_ext_check_inode(inode);
641 if (ret)
642 goto errout;
643
644 eh = ext_inode_hdr(inode);
645 ex = EXT_FIRST_EXTENT(eh);
646 if (ext4_blocks_count(es) > EXT4_MAX_BLOCK_FILE_PHYS ||
647 eh->eh_depth != 0 || le16_to_cpu(eh->eh_entries) > 1) {
648 ret = -EOPNOTSUPP;
649 goto errout;
650 }
651 if (eh->eh_entries == 0)
652 blk = len = start = end = 0;
653 else {
654 len = le16_to_cpu(ex->ee_len);
655 blk = ext4_ext_pblock(ex);
656 start = le32_to_cpu(ex->ee_block);
657 end = start + len - 1;
658 if (end >= EXT4_NDIR_BLOCKS) {
659 ret = -EOPNOTSUPP;
660 goto errout;
661 }
662 }
663
664 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
665 memset(ei->i_data, 0, sizeof(ei->i_data));
666 for (i = start; i <= end; i++)
667 ei->i_data[i] = cpu_to_le32(blk++);
668 ext4_mark_inode_dirty(handle, inode);
669errout:
670 ext4_journal_stop(handle);
671 up_write(&EXT4_I(inode)->i_data_sem);
672 return ret;
673}