Loading...
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 convenience 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 accumulate extra
47 * credit. But below we try to not accumulate 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 retval = ext4_datasem_ensure_credits(handle, inode, needed, needed, 0);
54 if (retval < 0)
55 goto err_out;
56 retval = ext4_ext_insert_extent(handle, inode, &path, &newext, 0);
57err_out:
58 up_write((&EXT4_I(inode)->i_data_sem));
59 ext4_free_ext_path(path);
60 lb->first_pblock = 0;
61 return retval;
62}
63
64static int update_extent_range(handle_t *handle, struct inode *inode,
65 ext4_fsblk_t pblock, struct migrate_struct *lb)
66{
67 int retval;
68 /*
69 * See if we can add on to the existing range (if it exists)
70 */
71 if (lb->first_pblock &&
72 (lb->last_pblock+1 == pblock) &&
73 (lb->last_block+1 == lb->curr_block)) {
74 lb->last_pblock = pblock;
75 lb->last_block = lb->curr_block;
76 lb->curr_block++;
77 return 0;
78 }
79 /*
80 * Start a new range.
81 */
82 retval = finish_range(handle, inode, lb);
83 lb->first_pblock = lb->last_pblock = pblock;
84 lb->first_block = lb->last_block = lb->curr_block;
85 lb->curr_block++;
86 return retval;
87}
88
89static int update_ind_extent_range(handle_t *handle, struct inode *inode,
90 ext4_fsblk_t pblock,
91 struct migrate_struct *lb)
92{
93 struct buffer_head *bh;
94 __le32 *i_data;
95 int i, retval = 0;
96 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
97
98 bh = ext4_sb_bread(inode->i_sb, pblock, 0);
99 if (IS_ERR(bh))
100 return PTR_ERR(bh);
101
102 i_data = (__le32 *)bh->b_data;
103 for (i = 0; i < max_entries; i++) {
104 if (i_data[i]) {
105 retval = update_extent_range(handle, inode,
106 le32_to_cpu(i_data[i]), lb);
107 if (retval)
108 break;
109 } else {
110 lb->curr_block++;
111 }
112 }
113 put_bh(bh);
114 return retval;
115
116}
117
118static int update_dind_extent_range(handle_t *handle, struct inode *inode,
119 ext4_fsblk_t pblock,
120 struct migrate_struct *lb)
121{
122 struct buffer_head *bh;
123 __le32 *i_data;
124 int i, retval = 0;
125 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
126
127 bh = ext4_sb_bread(inode->i_sb, pblock, 0);
128 if (IS_ERR(bh))
129 return PTR_ERR(bh);
130
131 i_data = (__le32 *)bh->b_data;
132 for (i = 0; i < max_entries; i++) {
133 if (i_data[i]) {
134 retval = update_ind_extent_range(handle, inode,
135 le32_to_cpu(i_data[i]), lb);
136 if (retval)
137 break;
138 } else {
139 /* Only update the file block number */
140 lb->curr_block += max_entries;
141 }
142 }
143 put_bh(bh);
144 return retval;
145
146}
147
148static int update_tind_extent_range(handle_t *handle, struct inode *inode,
149 ext4_fsblk_t pblock,
150 struct migrate_struct *lb)
151{
152 struct buffer_head *bh;
153 __le32 *i_data;
154 int i, retval = 0;
155 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
156
157 bh = ext4_sb_bread(inode->i_sb, pblock, 0);
158 if (IS_ERR(bh))
159 return PTR_ERR(bh);
160
161 i_data = (__le32 *)bh->b_data;
162 for (i = 0; i < max_entries; i++) {
163 if (i_data[i]) {
164 retval = update_dind_extent_range(handle, inode,
165 le32_to_cpu(i_data[i]), lb);
166 if (retval)
167 break;
168 } else {
169 /* Only update the file block number */
170 lb->curr_block += max_entries * max_entries;
171 }
172 }
173 put_bh(bh);
174 return retval;
175
176}
177
178static int free_dind_blocks(handle_t *handle,
179 struct inode *inode, __le32 i_data)
180{
181 int i;
182 __le32 *tmp_idata;
183 struct buffer_head *bh;
184 struct super_block *sb = inode->i_sb;
185 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
186 int err;
187
188 bh = ext4_sb_bread(sb, le32_to_cpu(i_data), 0);
189 if (IS_ERR(bh))
190 return PTR_ERR(bh);
191
192 tmp_idata = (__le32 *)bh->b_data;
193 for (i = 0; i < max_entries; i++) {
194 if (tmp_idata[i]) {
195 err = ext4_journal_ensure_credits(handle,
196 EXT4_RESERVE_TRANS_BLOCKS,
197 ext4_free_metadata_revoke_credits(sb, 1));
198 if (err < 0) {
199 put_bh(bh);
200 return err;
201 }
202 ext4_free_blocks(handle, inode, NULL,
203 le32_to_cpu(tmp_idata[i]), 1,
204 EXT4_FREE_BLOCKS_METADATA |
205 EXT4_FREE_BLOCKS_FORGET);
206 }
207 }
208 put_bh(bh);
209 err = ext4_journal_ensure_credits(handle, EXT4_RESERVE_TRANS_BLOCKS,
210 ext4_free_metadata_revoke_credits(sb, 1));
211 if (err < 0)
212 return err;
213 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
214 EXT4_FREE_BLOCKS_METADATA |
215 EXT4_FREE_BLOCKS_FORGET);
216 return 0;
217}
218
219static int free_tind_blocks(handle_t *handle,
220 struct inode *inode, __le32 i_data)
221{
222 int i, retval = 0;
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 retval = free_dind_blocks(handle,
235 inode, tmp_idata[i]);
236 if (retval) {
237 put_bh(bh);
238 return retval;
239 }
240 }
241 }
242 put_bh(bh);
243 retval = ext4_journal_ensure_credits(handle, EXT4_RESERVE_TRANS_BLOCKS,
244 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
245 if (retval < 0)
246 return retval;
247 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
248 EXT4_FREE_BLOCKS_METADATA |
249 EXT4_FREE_BLOCKS_FORGET);
250 return 0;
251}
252
253static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
254{
255 int retval;
256
257 /* ei->i_data[EXT4_IND_BLOCK] */
258 if (i_data[0]) {
259 retval = ext4_journal_ensure_credits(handle,
260 EXT4_RESERVE_TRANS_BLOCKS,
261 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
262 if (retval < 0)
263 return retval;
264 ext4_free_blocks(handle, inode, NULL,
265 le32_to_cpu(i_data[0]), 1,
266 EXT4_FREE_BLOCKS_METADATA |
267 EXT4_FREE_BLOCKS_FORGET);
268 }
269
270 /* ei->i_data[EXT4_DIND_BLOCK] */
271 if (i_data[1]) {
272 retval = free_dind_blocks(handle, inode, i_data[1]);
273 if (retval)
274 return retval;
275 }
276
277 /* ei->i_data[EXT4_TIND_BLOCK] */
278 if (i_data[2]) {
279 retval = free_tind_blocks(handle, inode, i_data[2]);
280 if (retval)
281 return retval;
282 }
283 return 0;
284}
285
286static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
287 struct inode *tmp_inode)
288{
289 int retval, retval2 = 0;
290 __le32 i_data[3];
291 struct ext4_inode_info *ei = EXT4_I(inode);
292 struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
293
294 /*
295 * One credit accounted for writing the
296 * i_data field of the original inode
297 */
298 retval = ext4_journal_ensure_credits(handle, 1, 0);
299 if (retval < 0)
300 goto err_out;
301
302 i_data[0] = ei->i_data[EXT4_IND_BLOCK];
303 i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
304 i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
305
306 down_write(&EXT4_I(inode)->i_data_sem);
307 /*
308 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
309 * happened after we started the migrate. We need to
310 * fail the migrate
311 */
312 if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) {
313 retval = -EAGAIN;
314 up_write(&EXT4_I(inode)->i_data_sem);
315 goto err_out;
316 } else
317 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
318 /*
319 * We have the extent map build with the tmp inode.
320 * Now copy the i_data across
321 */
322 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
323 memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
324
325 /*
326 * Update i_blocks with the new blocks that got
327 * allocated while adding extents for extent index
328 * blocks.
329 *
330 * While converting to extents we need not
331 * update the original inode i_blocks for extent blocks
332 * via quota APIs. The quota update happened via tmp_inode already.
333 */
334 spin_lock(&inode->i_lock);
335 inode->i_blocks += tmp_inode->i_blocks;
336 spin_unlock(&inode->i_lock);
337 up_write(&EXT4_I(inode)->i_data_sem);
338
339 /*
340 * We mark the inode dirty after, because we decrement the
341 * i_blocks when freeing the indirect meta-data blocks
342 */
343 retval = free_ind_block(handle, inode, i_data);
344 retval2 = ext4_mark_inode_dirty(handle, inode);
345 if (unlikely(retval2 && !retval))
346 retval = retval2;
347
348err_out:
349 return retval;
350}
351
352static int free_ext_idx(handle_t *handle, struct inode *inode,
353 struct ext4_extent_idx *ix)
354{
355 int i, retval = 0;
356 ext4_fsblk_t block;
357 struct buffer_head *bh;
358 struct ext4_extent_header *eh;
359
360 block = ext4_idx_pblock(ix);
361 bh = ext4_sb_bread(inode->i_sb, block, 0);
362 if (IS_ERR(bh))
363 return PTR_ERR(bh);
364
365 eh = (struct ext4_extent_header *)bh->b_data;
366 if (eh->eh_depth != 0) {
367 ix = EXT_FIRST_INDEX(eh);
368 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
369 retval = free_ext_idx(handle, inode, ix);
370 if (retval) {
371 put_bh(bh);
372 return retval;
373 }
374 }
375 }
376 put_bh(bh);
377 retval = ext4_journal_ensure_credits(handle, EXT4_RESERVE_TRANS_BLOCKS,
378 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
379 if (retval < 0)
380 return retval;
381 ext4_free_blocks(handle, inode, NULL, block, 1,
382 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
383 return 0;
384}
385
386/*
387 * Free the extent meta data blocks only
388 */
389static int free_ext_block(handle_t *handle, struct inode *inode)
390{
391 int i, retval = 0;
392 struct ext4_inode_info *ei = EXT4_I(inode);
393 struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
394 struct ext4_extent_idx *ix;
395 if (eh->eh_depth == 0)
396 /*
397 * No extra blocks allocated for extent meta data
398 */
399 return 0;
400 ix = EXT_FIRST_INDEX(eh);
401 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
402 retval = free_ext_idx(handle, inode, ix);
403 if (retval)
404 return retval;
405 }
406 return retval;
407}
408
409int ext4_ext_migrate(struct inode *inode)
410{
411 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
412 handle_t *handle;
413 int retval = 0, i;
414 __le32 *i_data;
415 struct ext4_inode_info *ei;
416 struct inode *tmp_inode = NULL;
417 struct migrate_struct lb;
418 unsigned long max_entries;
419 __u32 goal, tmp_csum_seed;
420 uid_t owner[2];
421
422 /*
423 * If the filesystem does not support extents, or the inode
424 * already is extent-based, error out.
425 */
426 if (!ext4_has_feature_extents(inode->i_sb) ||
427 ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
428 ext4_has_inline_data(inode))
429 return -EINVAL;
430
431 if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
432 /*
433 * don't migrate fast symlink
434 */
435 return retval;
436
437 percpu_down_write(&sbi->s_writepages_rwsem);
438
439 /*
440 * Worst case we can touch the allocation bitmaps and a block
441 * group descriptor block. We do need to worry about
442 * credits for modifying the quota inode.
443 */
444 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE,
445 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
446
447 if (IS_ERR(handle)) {
448 retval = PTR_ERR(handle);
449 goto out_unlock;
450 }
451 goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *
452 EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;
453 owner[0] = i_uid_read(inode);
454 owner[1] = i_gid_read(inode);
455 tmp_inode = ext4_new_inode(handle, d_inode(inode->i_sb->s_root),
456 S_IFREG, NULL, goal, owner, 0);
457 if (IS_ERR(tmp_inode)) {
458 retval = PTR_ERR(tmp_inode);
459 ext4_journal_stop(handle);
460 goto out_unlock;
461 }
462 /*
463 * Use the correct seed for checksum (i.e. the seed from 'inode'). This
464 * is so that the metadata blocks will have the correct checksum after
465 * the migration.
466 */
467 ei = EXT4_I(inode);
468 tmp_csum_seed = EXT4_I(tmp_inode)->i_csum_seed;
469 EXT4_I(tmp_inode)->i_csum_seed = ei->i_csum_seed;
470 i_size_write(tmp_inode, i_size_read(inode));
471 /*
472 * Set the i_nlink to zero so it will be deleted later
473 * when we drop inode reference.
474 */
475 clear_nlink(tmp_inode);
476
477 ext4_ext_tree_init(handle, tmp_inode);
478 ext4_journal_stop(handle);
479
480 /*
481 * start with one credit accounted for
482 * superblock modification.
483 *
484 * For the tmp_inode we already have committed the
485 * transaction that created the inode. Later as and
486 * when we add extents we extent the journal
487 */
488 /*
489 * Even though we take i_rwsem we can still cause block
490 * allocation via mmap write to holes. If we have allocated
491 * new blocks we fail migrate. New block allocation will
492 * clear EXT4_STATE_EXT_MIGRATE flag. The flag is updated
493 * with i_data_sem held to prevent racing with block
494 * allocation.
495 */
496 down_read(&EXT4_I(inode)->i_data_sem);
497 ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
498 up_read((&EXT4_I(inode)->i_data_sem));
499
500 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
501 if (IS_ERR(handle)) {
502 retval = PTR_ERR(handle);
503 goto out_tmp_inode;
504 }
505
506 i_data = ei->i_data;
507 memset(&lb, 0, sizeof(lb));
508
509 /* 32 bit block address 4 bytes */
510 max_entries = inode->i_sb->s_blocksize >> 2;
511 for (i = 0; i < EXT4_NDIR_BLOCKS; i++) {
512 if (i_data[i]) {
513 retval = update_extent_range(handle, tmp_inode,
514 le32_to_cpu(i_data[i]), &lb);
515 if (retval)
516 goto err_out;
517 } else
518 lb.curr_block++;
519 }
520 if (i_data[EXT4_IND_BLOCK]) {
521 retval = update_ind_extent_range(handle, tmp_inode,
522 le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb);
523 if (retval)
524 goto err_out;
525 } else
526 lb.curr_block += max_entries;
527 if (i_data[EXT4_DIND_BLOCK]) {
528 retval = update_dind_extent_range(handle, tmp_inode,
529 le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb);
530 if (retval)
531 goto err_out;
532 } else
533 lb.curr_block += max_entries * max_entries;
534 if (i_data[EXT4_TIND_BLOCK]) {
535 retval = update_tind_extent_range(handle, tmp_inode,
536 le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb);
537 if (retval)
538 goto err_out;
539 }
540 /*
541 * Build the last extent
542 */
543 retval = finish_range(handle, tmp_inode, &lb);
544err_out:
545 if (retval)
546 /*
547 * Failure case delete the extent information with the
548 * tmp_inode
549 */
550 free_ext_block(handle, tmp_inode);
551 else {
552 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
553 if (retval)
554 /*
555 * if we fail to swap inode data free the extent
556 * details of the tmp inode
557 */
558 free_ext_block(handle, tmp_inode);
559 }
560
561 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
562 retval = ext4_journal_ensure_credits(handle, 1, 0);
563 if (retval < 0)
564 goto out_stop;
565 /*
566 * Mark the tmp_inode as of size zero
567 */
568 i_size_write(tmp_inode, 0);
569
570 /*
571 * set the i_blocks count to zero
572 * so that the ext4_evict_inode() does the
573 * right job
574 *
575 * We don't need to take the i_lock because
576 * the inode is not visible to user space.
577 */
578 tmp_inode->i_blocks = 0;
579 EXT4_I(tmp_inode)->i_csum_seed = tmp_csum_seed;
580
581 /* Reset the extent details */
582 ext4_ext_tree_init(handle, tmp_inode);
583out_stop:
584 ext4_journal_stop(handle);
585out_tmp_inode:
586 unlock_new_inode(tmp_inode);
587 iput(tmp_inode);
588out_unlock:
589 percpu_up_write(&sbi->s_writepages_rwsem);
590 return retval;
591}
592
593/*
594 * Migrate a simple extent-based inode to use the i_blocks[] array
595 */
596int ext4_ind_migrate(struct inode *inode)
597{
598 struct ext4_extent_header *eh;
599 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
600 struct ext4_super_block *es = sbi->s_es;
601 struct ext4_inode_info *ei = EXT4_I(inode);
602 struct ext4_extent *ex;
603 unsigned int i, len;
604 ext4_lblk_t start, end;
605 ext4_fsblk_t blk;
606 handle_t *handle;
607 int ret, ret2 = 0;
608
609 if (!ext4_has_feature_extents(inode->i_sb) ||
610 (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
611 return -EINVAL;
612
613 if (ext4_has_feature_bigalloc(inode->i_sb))
614 return -EOPNOTSUPP;
615
616 /*
617 * In order to get correct extent info, force all delayed allocation
618 * blocks to be allocated, otherwise delayed allocation blocks may not
619 * be reflected and bypass the checks on extent header.
620 */
621 if (test_opt(inode->i_sb, DELALLOC))
622 ext4_alloc_da_blocks(inode);
623
624 percpu_down_write(&sbi->s_writepages_rwsem);
625
626 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
627 if (IS_ERR(handle)) {
628 ret = PTR_ERR(handle);
629 goto out_unlock;
630 }
631
632 down_write(&EXT4_I(inode)->i_data_sem);
633 ret = ext4_ext_check_inode(inode);
634 if (ret)
635 goto errout;
636
637 eh = ext_inode_hdr(inode);
638 ex = EXT_FIRST_EXTENT(eh);
639 if (ext4_blocks_count(es) > EXT4_MAX_BLOCK_FILE_PHYS ||
640 eh->eh_depth != 0 || le16_to_cpu(eh->eh_entries) > 1) {
641 ret = -EOPNOTSUPP;
642 goto errout;
643 }
644 if (eh->eh_entries == 0)
645 blk = len = start = end = 0;
646 else {
647 len = le16_to_cpu(ex->ee_len);
648 blk = ext4_ext_pblock(ex);
649 start = le32_to_cpu(ex->ee_block);
650 end = start + len - 1;
651 if (end >= EXT4_NDIR_BLOCKS) {
652 ret = -EOPNOTSUPP;
653 goto errout;
654 }
655 }
656
657 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
658 memset(ei->i_data, 0, sizeof(ei->i_data));
659 for (i = start; i <= end; i++)
660 ei->i_data[i] = cpu_to_le32(blk++);
661 ret2 = ext4_mark_inode_dirty(handle, inode);
662 if (unlikely(ret2 && !ret))
663 ret = ret2;
664errout:
665 ext4_journal_stop(handle);
666 up_write(&EXT4_I(inode)->i_data_sem);
667out_unlock:
668 percpu_up_write(&sbi->s_writepages_rwsem);
669 return ret;
670}
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}