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
18/*
19 * The contiguous blocks details which can be
20 * represented by a single extent
21 */
22struct migrate_struct {
23 ext4_lblk_t first_block, last_block, curr_block;
24 ext4_fsblk_t first_pblock, last_pblock;
25};
26
27static int finish_range(handle_t *handle, struct inode *inode,
28 struct migrate_struct *lb)
29
30{
31 int retval = 0, needed;
32 struct ext4_extent newext;
33 struct ext4_ext_path *path;
34 if (lb->first_pblock == 0)
35 return 0;
36
37 /* Add the extent to temp inode*/
38 newext.ee_block = cpu_to_le32(lb->first_block);
39 newext.ee_len = cpu_to_le16(lb->last_block - lb->first_block + 1);
40 ext4_ext_store_pblock(&newext, lb->first_pblock);
41 path = ext4_ext_find_extent(inode, lb->first_block, NULL);
42
43 if (IS_ERR(path)) {
44 retval = PTR_ERR(path);
45 path = NULL;
46 goto err_out;
47 }
48
49 /*
50 * Calculate the credit needed to inserting this extent
51 * Since we are doing this in loop we may accumalate extra
52 * credit. But below we try to not accumalate too much
53 * of them by restarting the journal.
54 */
55 needed = ext4_ext_calc_credits_for_single_extent(inode,
56 lb->last_block - lb->first_block + 1, path);
57
58 /*
59 * Make sure the credit we accumalated is not really high
60 */
61 if (needed && ext4_handle_has_enough_credits(handle,
62 EXT4_RESERVE_TRANS_BLOCKS)) {
63 retval = ext4_journal_restart(handle, needed);
64 if (retval)
65 goto err_out;
66 } else if (needed) {
67 retval = ext4_journal_extend(handle, needed);
68 if (retval) {
69 /*
70 * IF not able to extend the journal restart the journal
71 */
72 retval = ext4_journal_restart(handle, needed);
73 if (retval)
74 goto err_out;
75 }
76 }
77 retval = ext4_ext_insert_extent(handle, inode, path, &newext, 0);
78err_out:
79 if (path) {
80 ext4_ext_drop_refs(path);
81 kfree(path);
82 }
83 lb->first_pblock = 0;
84 return retval;
85}
86
87static int update_extent_range(handle_t *handle, struct inode *inode,
88 ext4_fsblk_t pblock, struct migrate_struct *lb)
89{
90 int retval;
91 /*
92 * See if we can add on to the existing range (if it exists)
93 */
94 if (lb->first_pblock &&
95 (lb->last_pblock+1 == pblock) &&
96 (lb->last_block+1 == lb->curr_block)) {
97 lb->last_pblock = pblock;
98 lb->last_block = lb->curr_block;
99 lb->curr_block++;
100 return 0;
101 }
102 /*
103 * Start a new range.
104 */
105 retval = finish_range(handle, inode, lb);
106 lb->first_pblock = lb->last_pblock = pblock;
107 lb->first_block = lb->last_block = lb->curr_block;
108 lb->curr_block++;
109 return retval;
110}
111
112static int update_ind_extent_range(handle_t *handle, struct inode *inode,
113 ext4_fsblk_t pblock,
114 struct migrate_struct *lb)
115{
116 struct buffer_head *bh;
117 __le32 *i_data;
118 int i, retval = 0;
119 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
120
121 bh = sb_bread(inode->i_sb, pblock);
122 if (!bh)
123 return -EIO;
124
125 i_data = (__le32 *)bh->b_data;
126 for (i = 0; i < max_entries; i++) {
127 if (i_data[i]) {
128 retval = update_extent_range(handle, inode,
129 le32_to_cpu(i_data[i]), lb);
130 if (retval)
131 break;
132 } else {
133 lb->curr_block++;
134 }
135 }
136 put_bh(bh);
137 return retval;
138
139}
140
141static int update_dind_extent_range(handle_t *handle, struct inode *inode,
142 ext4_fsblk_t pblock,
143 struct migrate_struct *lb)
144{
145 struct buffer_head *bh;
146 __le32 *i_data;
147 int i, retval = 0;
148 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
149
150 bh = sb_bread(inode->i_sb, pblock);
151 if (!bh)
152 return -EIO;
153
154 i_data = (__le32 *)bh->b_data;
155 for (i = 0; i < max_entries; i++) {
156 if (i_data[i]) {
157 retval = update_ind_extent_range(handle, inode,
158 le32_to_cpu(i_data[i]), lb);
159 if (retval)
160 break;
161 } else {
162 /* Only update the file block number */
163 lb->curr_block += max_entries;
164 }
165 }
166 put_bh(bh);
167 return retval;
168
169}
170
171static int update_tind_extent_range(handle_t *handle, struct inode *inode,
172 ext4_fsblk_t pblock,
173 struct migrate_struct *lb)
174{
175 struct buffer_head *bh;
176 __le32 *i_data;
177 int i, retval = 0;
178 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
179
180 bh = sb_bread(inode->i_sb, pblock);
181 if (!bh)
182 return -EIO;
183
184 i_data = (__le32 *)bh->b_data;
185 for (i = 0; i < max_entries; i++) {
186 if (i_data[i]) {
187 retval = update_dind_extent_range(handle, inode,
188 le32_to_cpu(i_data[i]), lb);
189 if (retval)
190 break;
191 } else {
192 /* Only update the file block number */
193 lb->curr_block += max_entries * max_entries;
194 }
195 }
196 put_bh(bh);
197 return retval;
198
199}
200
201static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
202{
203 int retval = 0, needed;
204
205 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
206 return 0;
207 /*
208 * We are freeing a blocks. During this we touch
209 * superblock, group descriptor and block bitmap.
210 * So allocate a credit of 3. We may update
211 * quota (user and group).
212 */
213 needed = 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
214
215 if (ext4_journal_extend(handle, needed) != 0)
216 retval = ext4_journal_restart(handle, needed);
217
218 return retval;
219}
220
221static int free_dind_blocks(handle_t *handle,
222 struct inode *inode, __le32 i_data)
223{
224 int i;
225 __le32 *tmp_idata;
226 struct buffer_head *bh;
227 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
228
229 bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
230 if (!bh)
231 return -EIO;
232
233 tmp_idata = (__le32 *)bh->b_data;
234 for (i = 0; i < max_entries; i++) {
235 if (tmp_idata[i]) {
236 extend_credit_for_blkdel(handle, inode);
237 ext4_free_blocks(handle, inode, NULL,
238 le32_to_cpu(tmp_idata[i]), 1,
239 EXT4_FREE_BLOCKS_METADATA |
240 EXT4_FREE_BLOCKS_FORGET);
241 }
242 }
243 put_bh(bh);
244 extend_credit_for_blkdel(handle, inode);
245 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
246 EXT4_FREE_BLOCKS_METADATA |
247 EXT4_FREE_BLOCKS_FORGET);
248 return 0;
249}
250
251static int free_tind_blocks(handle_t *handle,
252 struct inode *inode, __le32 i_data)
253{
254 int i, retval = 0;
255 __le32 *tmp_idata;
256 struct buffer_head *bh;
257 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
258
259 bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
260 if (!bh)
261 return -EIO;
262
263 tmp_idata = (__le32 *)bh->b_data;
264 for (i = 0; i < max_entries; i++) {
265 if (tmp_idata[i]) {
266 retval = free_dind_blocks(handle,
267 inode, tmp_idata[i]);
268 if (retval) {
269 put_bh(bh);
270 return retval;
271 }
272 }
273 }
274 put_bh(bh);
275 extend_credit_for_blkdel(handle, inode);
276 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
277 EXT4_FREE_BLOCKS_METADATA |
278 EXT4_FREE_BLOCKS_FORGET);
279 return 0;
280}
281
282static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
283{
284 int retval;
285
286 /* ei->i_data[EXT4_IND_BLOCK] */
287 if (i_data[0]) {
288 extend_credit_for_blkdel(handle, inode);
289 ext4_free_blocks(handle, inode, NULL,
290 le32_to_cpu(i_data[0]), 1,
291 EXT4_FREE_BLOCKS_METADATA |
292 EXT4_FREE_BLOCKS_FORGET);
293 }
294
295 /* ei->i_data[EXT4_DIND_BLOCK] */
296 if (i_data[1]) {
297 retval = free_dind_blocks(handle, inode, i_data[1]);
298 if (retval)
299 return retval;
300 }
301
302 /* ei->i_data[EXT4_TIND_BLOCK] */
303 if (i_data[2]) {
304 retval = free_tind_blocks(handle, inode, i_data[2]);
305 if (retval)
306 return retval;
307 }
308 return 0;
309}
310
311static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
312 struct inode *tmp_inode)
313{
314 int retval;
315 __le32 i_data[3];
316 struct ext4_inode_info *ei = EXT4_I(inode);
317 struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
318
319 /*
320 * One credit accounted for writing the
321 * i_data field of the original inode
322 */
323 retval = ext4_journal_extend(handle, 1);
324 if (retval) {
325 retval = ext4_journal_restart(handle, 1);
326 if (retval)
327 goto err_out;
328 }
329
330 i_data[0] = ei->i_data[EXT4_IND_BLOCK];
331 i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
332 i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
333
334 down_write(&EXT4_I(inode)->i_data_sem);
335 /*
336 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
337 * happened after we started the migrate. We need to
338 * fail the migrate
339 */
340 if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) {
341 retval = -EAGAIN;
342 up_write(&EXT4_I(inode)->i_data_sem);
343 goto err_out;
344 } else
345 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
346 /*
347 * We have the extent map build with the tmp inode.
348 * Now copy the i_data across
349 */
350 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
351 memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
352
353 /*
354 * Update i_blocks with the new blocks that got
355 * allocated while adding extents for extent index
356 * blocks.
357 *
358 * While converting to extents we need not
359 * update the orignal inode i_blocks for extent blocks
360 * via quota APIs. The quota update happened via tmp_inode already.
361 */
362 spin_lock(&inode->i_lock);
363 inode->i_blocks += tmp_inode->i_blocks;
364 spin_unlock(&inode->i_lock);
365 up_write(&EXT4_I(inode)->i_data_sem);
366
367 /*
368 * We mark the inode dirty after, because we decrement the
369 * i_blocks when freeing the indirect meta-data blocks
370 */
371 retval = free_ind_block(handle, inode, i_data);
372 ext4_mark_inode_dirty(handle, inode);
373
374err_out:
375 return retval;
376}
377
378static int free_ext_idx(handle_t *handle, struct inode *inode,
379 struct ext4_extent_idx *ix)
380{
381 int i, retval = 0;
382 ext4_fsblk_t block;
383 struct buffer_head *bh;
384 struct ext4_extent_header *eh;
385
386 block = ext4_idx_pblock(ix);
387 bh = sb_bread(inode->i_sb, block);
388 if (!bh)
389 return -EIO;
390
391 eh = (struct ext4_extent_header *)bh->b_data;
392 if (eh->eh_depth != 0) {
393 ix = EXT_FIRST_INDEX(eh);
394 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
395 retval = free_ext_idx(handle, inode, ix);
396 if (retval)
397 break;
398 }
399 }
400 put_bh(bh);
401 extend_credit_for_blkdel(handle, inode);
402 ext4_free_blocks(handle, inode, NULL, block, 1,
403 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
404 return retval;
405}
406
407/*
408 * Free the extent meta data blocks only
409 */
410static int free_ext_block(handle_t *handle, struct inode *inode)
411{
412 int i, retval = 0;
413 struct ext4_inode_info *ei = EXT4_I(inode);
414 struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
415 struct ext4_extent_idx *ix;
416 if (eh->eh_depth == 0)
417 /*
418 * No extra blocks allocated for extent meta data
419 */
420 return 0;
421 ix = EXT_FIRST_INDEX(eh);
422 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
423 retval = free_ext_idx(handle, inode, ix);
424 if (retval)
425 return retval;
426 }
427 return retval;
428
429}
430
431int ext4_ext_migrate(struct inode *inode)
432{
433 handle_t *handle;
434 int retval = 0, i;
435 __le32 *i_data;
436 struct ext4_inode_info *ei;
437 struct inode *tmp_inode = NULL;
438 struct migrate_struct lb;
439 unsigned long max_entries;
440 __u32 goal;
441 uid_t owner[2];
442
443 /*
444 * If the filesystem does not support extents, or the inode
445 * already is extent-based, error out.
446 */
447 if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
448 EXT4_FEATURE_INCOMPAT_EXTENTS) ||
449 (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
450 return -EINVAL;
451
452 if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
453 /*
454 * don't migrate fast symlink
455 */
456 return retval;
457
458 handle = ext4_journal_start(inode,
459 EXT4_DATA_TRANS_BLOCKS(inode->i_sb) +
460 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
461 EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)
462 + 1);
463 if (IS_ERR(handle)) {
464 retval = PTR_ERR(handle);
465 return retval;
466 }
467 goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *
468 EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;
469 owner[0] = i_uid_read(inode);
470 owner[1] = i_gid_read(inode);
471 tmp_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
472 S_IFREG, NULL, goal, owner);
473 if (IS_ERR(tmp_inode)) {
474 retval = PTR_ERR(tmp_inode);
475 ext4_journal_stop(handle);
476 return retval;
477 }
478 i_size_write(tmp_inode, i_size_read(inode));
479 /*
480 * Set the i_nlink to zero so it will be deleted later
481 * when we drop inode reference.
482 */
483 clear_nlink(tmp_inode);
484
485 ext4_ext_tree_init(handle, tmp_inode);
486 ext4_orphan_add(handle, tmp_inode);
487 ext4_journal_stop(handle);
488
489 /*
490 * start with one credit accounted for
491 * superblock modification.
492 *
493 * For the tmp_inode we already have committed the
494 * trascation that created the inode. Later as and
495 * when we add extents we extent the journal
496 */
497 /*
498 * Even though we take i_mutex we can still cause block
499 * allocation via mmap write to holes. If we have allocated
500 * new blocks we fail migrate. New block allocation will
501 * clear EXT4_STATE_EXT_MIGRATE flag. The flag is updated
502 * with i_data_sem held to prevent racing with block
503 * allocation.
504 */
505 down_read((&EXT4_I(inode)->i_data_sem));
506 ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
507 up_read((&EXT4_I(inode)->i_data_sem));
508
509 handle = ext4_journal_start(inode, 1);
510 if (IS_ERR(handle)) {
511 /*
512 * It is impossible to update on-disk structures without
513 * a handle, so just rollback in-core changes and live other
514 * work to orphan_list_cleanup()
515 */
516 ext4_orphan_del(NULL, tmp_inode);
517 retval = PTR_ERR(handle);
518 goto out;
519 }
520
521 ei = EXT4_I(inode);
522 i_data = ei->i_data;
523 memset(&lb, 0, sizeof(lb));
524
525 /* 32 bit block address 4 bytes */
526 max_entries = inode->i_sb->s_blocksize >> 2;
527 for (i = 0; i < EXT4_NDIR_BLOCKS; i++) {
528 if (i_data[i]) {
529 retval = update_extent_range(handle, tmp_inode,
530 le32_to_cpu(i_data[i]), &lb);
531 if (retval)
532 goto err_out;
533 } else
534 lb.curr_block++;
535 }
536 if (i_data[EXT4_IND_BLOCK]) {
537 retval = update_ind_extent_range(handle, tmp_inode,
538 le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb);
539 if (retval)
540 goto err_out;
541 } else
542 lb.curr_block += max_entries;
543 if (i_data[EXT4_DIND_BLOCK]) {
544 retval = update_dind_extent_range(handle, tmp_inode,
545 le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb);
546 if (retval)
547 goto err_out;
548 } else
549 lb.curr_block += max_entries * max_entries;
550 if (i_data[EXT4_TIND_BLOCK]) {
551 retval = update_tind_extent_range(handle, tmp_inode,
552 le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb);
553 if (retval)
554 goto err_out;
555 }
556 /*
557 * Build the last extent
558 */
559 retval = finish_range(handle, tmp_inode, &lb);
560err_out:
561 if (retval)
562 /*
563 * Failure case delete the extent information with the
564 * tmp_inode
565 */
566 free_ext_block(handle, tmp_inode);
567 else {
568 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
569 if (retval)
570 /*
571 * if we fail to swap inode data free the extent
572 * details of the tmp inode
573 */
574 free_ext_block(handle, tmp_inode);
575 }
576
577 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
578 if (ext4_journal_extend(handle, 1) != 0)
579 ext4_journal_restart(handle, 1);
580
581 /*
582 * Mark the tmp_inode as of size zero
583 */
584 i_size_write(tmp_inode, 0);
585
586 /*
587 * set the i_blocks count to zero
588 * so that the ext4_delete_inode does the
589 * right job
590 *
591 * We don't need to take the i_lock because
592 * the inode is not visible to user space.
593 */
594 tmp_inode->i_blocks = 0;
595
596 /* Reset the extent details */
597 ext4_ext_tree_init(handle, tmp_inode);
598 ext4_journal_stop(handle);
599out:
600 unlock_new_inode(tmp_inode);
601 iput(tmp_inode);
602
603 return retval;
604}
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 path = ext4_ext_find_extent(inode, lb->first_block, NULL, 0);
43
44 if (IS_ERR(path)) {
45 retval = PTR_ERR(path);
46 path = NULL;
47 goto err_out;
48 }
49
50 /*
51 * Calculate the credit needed to inserting this extent
52 * Since we are doing this in loop we may accumalate extra
53 * credit. But below we try to not accumalate too much
54 * of them by restarting the journal.
55 */
56 needed = ext4_ext_calc_credits_for_single_extent(inode,
57 lb->last_block - lb->first_block + 1, path);
58
59 /*
60 * Make sure the credit we accumalated is not really high
61 */
62 if (needed && ext4_handle_has_enough_credits(handle,
63 EXT4_RESERVE_TRANS_BLOCKS)) {
64 retval = ext4_journal_restart(handle, needed);
65 if (retval)
66 goto err_out;
67 } else if (needed) {
68 retval = ext4_journal_extend(handle, needed);
69 if (retval) {
70 /*
71 * IF not able to extend the journal restart the journal
72 */
73 retval = ext4_journal_restart(handle, needed);
74 if (retval)
75 goto err_out;
76 }
77 }
78 retval = ext4_ext_insert_extent(handle, inode, path, &newext, 0);
79err_out:
80 if (path) {
81 ext4_ext_drop_refs(path);
82 kfree(path);
83 }
84 lb->first_pblock = 0;
85 return retval;
86}
87
88static int update_extent_range(handle_t *handle, struct inode *inode,
89 ext4_fsblk_t pblock, struct migrate_struct *lb)
90{
91 int retval;
92 /*
93 * See if we can add on to the existing range (if it exists)
94 */
95 if (lb->first_pblock &&
96 (lb->last_pblock+1 == pblock) &&
97 (lb->last_block+1 == lb->curr_block)) {
98 lb->last_pblock = pblock;
99 lb->last_block = lb->curr_block;
100 lb->curr_block++;
101 return 0;
102 }
103 /*
104 * Start a new range.
105 */
106 retval = finish_range(handle, inode, lb);
107 lb->first_pblock = lb->last_pblock = pblock;
108 lb->first_block = lb->last_block = lb->curr_block;
109 lb->curr_block++;
110 return retval;
111}
112
113static int update_ind_extent_range(handle_t *handle, struct inode *inode,
114 ext4_fsblk_t pblock,
115 struct migrate_struct *lb)
116{
117 struct buffer_head *bh;
118 __le32 *i_data;
119 int i, retval = 0;
120 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
121
122 bh = sb_bread(inode->i_sb, pblock);
123 if (!bh)
124 return -EIO;
125
126 i_data = (__le32 *)bh->b_data;
127 for (i = 0; i < max_entries; i++) {
128 if (i_data[i]) {
129 retval = update_extent_range(handle, inode,
130 le32_to_cpu(i_data[i]), lb);
131 if (retval)
132 break;
133 } else {
134 lb->curr_block++;
135 }
136 }
137 put_bh(bh);
138 return retval;
139
140}
141
142static int update_dind_extent_range(handle_t *handle, struct inode *inode,
143 ext4_fsblk_t pblock,
144 struct migrate_struct *lb)
145{
146 struct buffer_head *bh;
147 __le32 *i_data;
148 int i, retval = 0;
149 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
150
151 bh = sb_bread(inode->i_sb, pblock);
152 if (!bh)
153 return -EIO;
154
155 i_data = (__le32 *)bh->b_data;
156 for (i = 0; i < max_entries; i++) {
157 if (i_data[i]) {
158 retval = update_ind_extent_range(handle, inode,
159 le32_to_cpu(i_data[i]), lb);
160 if (retval)
161 break;
162 } else {
163 /* Only update the file block number */
164 lb->curr_block += max_entries;
165 }
166 }
167 put_bh(bh);
168 return retval;
169
170}
171
172static int update_tind_extent_range(handle_t *handle, struct inode *inode,
173 ext4_fsblk_t pblock,
174 struct migrate_struct *lb)
175{
176 struct buffer_head *bh;
177 __le32 *i_data;
178 int i, retval = 0;
179 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
180
181 bh = sb_bread(inode->i_sb, pblock);
182 if (!bh)
183 return -EIO;
184
185 i_data = (__le32 *)bh->b_data;
186 for (i = 0; i < max_entries; i++) {
187 if (i_data[i]) {
188 retval = update_dind_extent_range(handle, inode,
189 le32_to_cpu(i_data[i]), lb);
190 if (retval)
191 break;
192 } else {
193 /* Only update the file block number */
194 lb->curr_block += max_entries * max_entries;
195 }
196 }
197 put_bh(bh);
198 return retval;
199
200}
201
202static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
203{
204 int retval = 0, needed;
205
206 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
207 return 0;
208 /*
209 * We are freeing a blocks. During this we touch
210 * superblock, group descriptor and block bitmap.
211 * So allocate a credit of 3. We may update
212 * quota (user and group).
213 */
214 needed = 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
215
216 if (ext4_journal_extend(handle, needed) != 0)
217 retval = ext4_journal_restart(handle, needed);
218
219 return retval;
220}
221
222static int free_dind_blocks(handle_t *handle,
223 struct inode *inode, __le32 i_data)
224{
225 int i;
226 __le32 *tmp_idata;
227 struct buffer_head *bh;
228 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
229
230 bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
231 if (!bh)
232 return -EIO;
233
234 tmp_idata = (__le32 *)bh->b_data;
235 for (i = 0; i < max_entries; i++) {
236 if (tmp_idata[i]) {
237 extend_credit_for_blkdel(handle, inode);
238 ext4_free_blocks(handle, inode, NULL,
239 le32_to_cpu(tmp_idata[i]), 1,
240 EXT4_FREE_BLOCKS_METADATA |
241 EXT4_FREE_BLOCKS_FORGET);
242 }
243 }
244 put_bh(bh);
245 extend_credit_for_blkdel(handle, inode);
246 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
247 EXT4_FREE_BLOCKS_METADATA |
248 EXT4_FREE_BLOCKS_FORGET);
249 return 0;
250}
251
252static int free_tind_blocks(handle_t *handle,
253 struct inode *inode, __le32 i_data)
254{
255 int i, retval = 0;
256 __le32 *tmp_idata;
257 struct buffer_head *bh;
258 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
259
260 bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
261 if (!bh)
262 return -EIO;
263
264 tmp_idata = (__le32 *)bh->b_data;
265 for (i = 0; i < max_entries; i++) {
266 if (tmp_idata[i]) {
267 retval = free_dind_blocks(handle,
268 inode, tmp_idata[i]);
269 if (retval) {
270 put_bh(bh);
271 return retval;
272 }
273 }
274 }
275 put_bh(bh);
276 extend_credit_for_blkdel(handle, inode);
277 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
278 EXT4_FREE_BLOCKS_METADATA |
279 EXT4_FREE_BLOCKS_FORGET);
280 return 0;
281}
282
283static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
284{
285 int retval;
286
287 /* ei->i_data[EXT4_IND_BLOCK] */
288 if (i_data[0]) {
289 extend_credit_for_blkdel(handle, inode);
290 ext4_free_blocks(handle, inode, NULL,
291 le32_to_cpu(i_data[0]), 1,
292 EXT4_FREE_BLOCKS_METADATA |
293 EXT4_FREE_BLOCKS_FORGET);
294 }
295
296 /* ei->i_data[EXT4_DIND_BLOCK] */
297 if (i_data[1]) {
298 retval = free_dind_blocks(handle, inode, i_data[1]);
299 if (retval)
300 return retval;
301 }
302
303 /* ei->i_data[EXT4_TIND_BLOCK] */
304 if (i_data[2]) {
305 retval = free_tind_blocks(handle, inode, i_data[2]);
306 if (retval)
307 return retval;
308 }
309 return 0;
310}
311
312static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
313 struct inode *tmp_inode)
314{
315 int retval;
316 __le32 i_data[3];
317 struct ext4_inode_info *ei = EXT4_I(inode);
318 struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
319
320 /*
321 * One credit accounted for writing the
322 * i_data field of the original inode
323 */
324 retval = ext4_journal_extend(handle, 1);
325 if (retval) {
326 retval = ext4_journal_restart(handle, 1);
327 if (retval)
328 goto err_out;
329 }
330
331 i_data[0] = ei->i_data[EXT4_IND_BLOCK];
332 i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
333 i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
334
335 down_write(&EXT4_I(inode)->i_data_sem);
336 /*
337 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
338 * happened after we started the migrate. We need to
339 * fail the migrate
340 */
341 if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) {
342 retval = -EAGAIN;
343 up_write(&EXT4_I(inode)->i_data_sem);
344 goto err_out;
345 } else
346 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
347 /*
348 * We have the extent map build with the tmp inode.
349 * Now copy the i_data across
350 */
351 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
352 memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
353
354 /*
355 * Update i_blocks with the new blocks that got
356 * allocated while adding extents for extent index
357 * blocks.
358 *
359 * While converting to extents we need not
360 * update the orignal inode i_blocks for extent blocks
361 * via quota APIs. The quota update happened via tmp_inode already.
362 */
363 spin_lock(&inode->i_lock);
364 inode->i_blocks += tmp_inode->i_blocks;
365 spin_unlock(&inode->i_lock);
366 up_write(&EXT4_I(inode)->i_data_sem);
367
368 /*
369 * We mark the inode dirty after, because we decrement the
370 * i_blocks when freeing the indirect meta-data blocks
371 */
372 retval = free_ind_block(handle, inode, i_data);
373 ext4_mark_inode_dirty(handle, inode);
374
375err_out:
376 return retval;
377}
378
379static int free_ext_idx(handle_t *handle, struct inode *inode,
380 struct ext4_extent_idx *ix)
381{
382 int i, retval = 0;
383 ext4_fsblk_t block;
384 struct buffer_head *bh;
385 struct ext4_extent_header *eh;
386
387 block = ext4_idx_pblock(ix);
388 bh = sb_bread(inode->i_sb, block);
389 if (!bh)
390 return -EIO;
391
392 eh = (struct ext4_extent_header *)bh->b_data;
393 if (eh->eh_depth != 0) {
394 ix = EXT_FIRST_INDEX(eh);
395 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
396 retval = free_ext_idx(handle, inode, ix);
397 if (retval)
398 break;
399 }
400 }
401 put_bh(bh);
402 extend_credit_for_blkdel(handle, inode);
403 ext4_free_blocks(handle, inode, NULL, block, 1,
404 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
405 return retval;
406}
407
408/*
409 * Free the extent meta data blocks only
410 */
411static int free_ext_block(handle_t *handle, struct inode *inode)
412{
413 int i, retval = 0;
414 struct ext4_inode_info *ei = EXT4_I(inode);
415 struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
416 struct ext4_extent_idx *ix;
417 if (eh->eh_depth == 0)
418 /*
419 * No extra blocks allocated for extent meta data
420 */
421 return 0;
422 ix = EXT_FIRST_INDEX(eh);
423 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
424 retval = free_ext_idx(handle, inode, ix);
425 if (retval)
426 return retval;
427 }
428 return retval;
429}
430
431int ext4_ext_migrate(struct inode *inode)
432{
433 handle_t *handle;
434 int retval = 0, i;
435 __le32 *i_data;
436 struct ext4_inode_info *ei;
437 struct inode *tmp_inode = NULL;
438 struct migrate_struct lb;
439 unsigned long max_entries;
440 __u32 goal;
441 uid_t owner[2];
442
443 /*
444 * If the filesystem does not support extents, or the inode
445 * already is extent-based, error out.
446 */
447 if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
448 EXT4_FEATURE_INCOMPAT_EXTENTS) ||
449 (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
450 return -EINVAL;
451
452 if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
453 /*
454 * don't migrate fast symlink
455 */
456 return retval;
457
458 /*
459 * Worst case we can touch the allocation bitmaps, a bgd
460 * block, and a block to link in the orphan list. We do need
461 * need to worry about credits for modifying the quota inode.
462 */
463 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE,
464 4 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
465
466 if (IS_ERR(handle)) {
467 retval = PTR_ERR(handle);
468 return retval;
469 }
470 goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *
471 EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;
472 owner[0] = i_uid_read(inode);
473 owner[1] = i_gid_read(inode);
474 tmp_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
475 S_IFREG, NULL, goal, owner);
476 if (IS_ERR(tmp_inode)) {
477 retval = PTR_ERR(tmp_inode);
478 ext4_journal_stop(handle);
479 return retval;
480 }
481 i_size_write(tmp_inode, i_size_read(inode));
482 /*
483 * Set the i_nlink to zero so it will be deleted later
484 * when we drop inode reference.
485 */
486 clear_nlink(tmp_inode);
487
488 ext4_ext_tree_init(handle, tmp_inode);
489 ext4_orphan_add(handle, tmp_inode);
490 ext4_journal_stop(handle);
491
492 /*
493 * start with one credit accounted for
494 * superblock modification.
495 *
496 * For the tmp_inode we already have committed the
497 * transaction that created the inode. Later as and
498 * when we add extents we extent the journal
499 */
500 /*
501 * Even though we take i_mutex we can still cause block
502 * allocation via mmap write to holes. If we have allocated
503 * new blocks we fail migrate. New block allocation will
504 * clear EXT4_STATE_EXT_MIGRATE flag. The flag is updated
505 * with i_data_sem held to prevent racing with block
506 * allocation.
507 */
508 down_read((&EXT4_I(inode)->i_data_sem));
509 ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
510 up_read((&EXT4_I(inode)->i_data_sem));
511
512 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
513 if (IS_ERR(handle)) {
514 /*
515 * It is impossible to update on-disk structures without
516 * a handle, so just rollback in-core changes and live other
517 * work to orphan_list_cleanup()
518 */
519 ext4_orphan_del(NULL, tmp_inode);
520 retval = PTR_ERR(handle);
521 goto out;
522 }
523
524 ei = EXT4_I(inode);
525 i_data = ei->i_data;
526 memset(&lb, 0, sizeof(lb));
527
528 /* 32 bit block address 4 bytes */
529 max_entries = inode->i_sb->s_blocksize >> 2;
530 for (i = 0; i < EXT4_NDIR_BLOCKS; i++) {
531 if (i_data[i]) {
532 retval = update_extent_range(handle, tmp_inode,
533 le32_to_cpu(i_data[i]), &lb);
534 if (retval)
535 goto err_out;
536 } else
537 lb.curr_block++;
538 }
539 if (i_data[EXT4_IND_BLOCK]) {
540 retval = update_ind_extent_range(handle, tmp_inode,
541 le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb);
542 if (retval)
543 goto err_out;
544 } else
545 lb.curr_block += max_entries;
546 if (i_data[EXT4_DIND_BLOCK]) {
547 retval = update_dind_extent_range(handle, tmp_inode,
548 le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb);
549 if (retval)
550 goto err_out;
551 } else
552 lb.curr_block += max_entries * max_entries;
553 if (i_data[EXT4_TIND_BLOCK]) {
554 retval = update_tind_extent_range(handle, tmp_inode,
555 le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb);
556 if (retval)
557 goto err_out;
558 }
559 /*
560 * Build the last extent
561 */
562 retval = finish_range(handle, tmp_inode, &lb);
563err_out:
564 if (retval)
565 /*
566 * Failure case delete the extent information with the
567 * tmp_inode
568 */
569 free_ext_block(handle, tmp_inode);
570 else {
571 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
572 if (retval)
573 /*
574 * if we fail to swap inode data free the extent
575 * details of the tmp inode
576 */
577 free_ext_block(handle, tmp_inode);
578 }
579
580 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
581 if (ext4_journal_extend(handle, 1) != 0)
582 ext4_journal_restart(handle, 1);
583
584 /*
585 * Mark the tmp_inode as of size zero
586 */
587 i_size_write(tmp_inode, 0);
588
589 /*
590 * set the i_blocks count to zero
591 * so that the ext4_delete_inode does the
592 * right job
593 *
594 * We don't need to take the i_lock because
595 * the inode is not visible to user space.
596 */
597 tmp_inode->i_blocks = 0;
598
599 /* Reset the extent details */
600 ext4_ext_tree_init(handle, tmp_inode);
601 ext4_journal_stop(handle);
602out:
603 unlock_new_inode(tmp_inode);
604 iput(tmp_inode);
605
606 return retval;
607}
608
609/*
610 * Migrate a simple extent-based inode to use the i_blocks[] array
611 */
612int ext4_ind_migrate(struct inode *inode)
613{
614 struct ext4_extent_header *eh;
615 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
616 struct ext4_inode_info *ei = EXT4_I(inode);
617 struct ext4_extent *ex;
618 unsigned int i, len;
619 ext4_fsblk_t blk;
620 handle_t *handle;
621 int ret;
622
623 if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
624 EXT4_FEATURE_INCOMPAT_EXTENTS) ||
625 (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
626 return -EINVAL;
627
628 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
629 EXT4_FEATURE_RO_COMPAT_BIGALLOC))
630 return -EOPNOTSUPP;
631
632 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
633 if (IS_ERR(handle))
634 return PTR_ERR(handle);
635
636 down_write(&EXT4_I(inode)->i_data_sem);
637 ret = ext4_ext_check_inode(inode);
638 if (ret)
639 goto errout;
640
641 eh = ext_inode_hdr(inode);
642 ex = EXT_FIRST_EXTENT(eh);
643 if (ext4_blocks_count(es) > EXT4_MAX_BLOCK_FILE_PHYS ||
644 eh->eh_depth != 0 || le16_to_cpu(eh->eh_entries) > 1) {
645 ret = -EOPNOTSUPP;
646 goto errout;
647 }
648 if (eh->eh_entries == 0)
649 blk = len = 0;
650 else {
651 len = le16_to_cpu(ex->ee_len);
652 blk = ext4_ext_pblock(ex);
653 if (len > EXT4_NDIR_BLOCKS) {
654 ret = -EOPNOTSUPP;
655 goto errout;
656 }
657 }
658
659 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
660 memset(ei->i_data, 0, sizeof(ei->i_data));
661 for (i=0; i < len; i++)
662 ei->i_data[i] = cpu_to_le32(blk++);
663 ext4_mark_inode_dirty(handle, inode);
664errout:
665 ext4_journal_stop(handle);
666 up_write(&EXT4_I(inode)->i_data_sem);
667 return ret;
668}