ext4: teach ext4_ext_find_extent() to realloc path if necessary
[firefly-linux-kernel-4.4.55.git] / fs / ext4 / migrate.c
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  */
23 struct migrate_struct {
24         ext4_lblk_t first_block, last_block, curr_block;
25         ext4_fsblk_t first_pblock, last_pblock;
26 };
27
28 static 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_ext_find_extent(inode, lb->first_block, NULL, 0);
45
46         if (IS_ERR(path)) {
47                 retval = PTR_ERR(path);
48                 path = NULL;
49                 goto err_out;
50         }
51
52         /*
53          * Calculate the credit needed to inserting this extent
54          * Since we are doing this in loop we may accumalate extra
55          * credit. But below we try to not accumalate too much
56          * of them by restarting the journal.
57          */
58         needed = ext4_ext_calc_credits_for_single_extent(inode,
59                     lb->last_block - lb->first_block + 1, path);
60
61         /*
62          * Make sure the credit we accumalated is not really high
63          */
64         if (needed && ext4_handle_has_enough_credits(handle,
65                                                 EXT4_RESERVE_TRANS_BLOCKS)) {
66                 up_write((&EXT4_I(inode)->i_data_sem));
67                 retval = ext4_journal_restart(handle, needed);
68                 down_write((&EXT4_I(inode)->i_data_sem));
69                 if (retval)
70                         goto err_out;
71         } else if (needed) {
72                 retval = ext4_journal_extend(handle, needed);
73                 if (retval) {
74                         /*
75                          * IF not able to extend the journal restart the journal
76                          */
77                         up_write((&EXT4_I(inode)->i_data_sem));
78                         retval = ext4_journal_restart(handle, needed);
79                         down_write((&EXT4_I(inode)->i_data_sem));
80                         if (retval)
81                                 goto err_out;
82                 }
83         }
84         retval = ext4_ext_insert_extent(handle, inode, &path, &newext, 0);
85 err_out:
86         up_write((&EXT4_I(inode)->i_data_sem));
87         ext4_ext_drop_refs(path);
88         kfree(path);
89         lb->first_pblock = 0;
90         return retval;
91 }
92
93 static int update_extent_range(handle_t *handle, struct inode *inode,
94                                ext4_fsblk_t pblock, struct migrate_struct *lb)
95 {
96         int retval;
97         /*
98          * See if we can add on to the existing range (if it exists)
99          */
100         if (lb->first_pblock &&
101                 (lb->last_pblock+1 == pblock) &&
102                 (lb->last_block+1 == lb->curr_block)) {
103                 lb->last_pblock = pblock;
104                 lb->last_block = lb->curr_block;
105                 lb->curr_block++;
106                 return 0;
107         }
108         /*
109          * Start a new range.
110          */
111         retval = finish_range(handle, inode, lb);
112         lb->first_pblock = lb->last_pblock = pblock;
113         lb->first_block = lb->last_block = lb->curr_block;
114         lb->curr_block++;
115         return retval;
116 }
117
118 static int update_ind_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 = sb_bread(inode->i_sb, pblock);
128         if (!bh)
129                 return -EIO;
130
131         i_data = (__le32 *)bh->b_data;
132         for (i = 0; i < max_entries; i++) {
133                 if (i_data[i]) {
134                         retval = update_extent_range(handle, inode,
135                                                 le32_to_cpu(i_data[i]), lb);
136                         if (retval)
137                                 break;
138                 } else {
139                         lb->curr_block++;
140                 }
141         }
142         put_bh(bh);
143         return retval;
144
145 }
146
147 static int update_dind_extent_range(handle_t *handle, struct inode *inode,
148                                     ext4_fsblk_t pblock,
149                                     struct migrate_struct *lb)
150 {
151         struct buffer_head *bh;
152         __le32 *i_data;
153         int i, retval = 0;
154         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
155
156         bh = sb_bread(inode->i_sb, pblock);
157         if (!bh)
158                 return -EIO;
159
160         i_data = (__le32 *)bh->b_data;
161         for (i = 0; i < max_entries; i++) {
162                 if (i_data[i]) {
163                         retval = update_ind_extent_range(handle, inode,
164                                                 le32_to_cpu(i_data[i]), lb);
165                         if (retval)
166                                 break;
167                 } else {
168                         /* Only update the file block number */
169                         lb->curr_block += max_entries;
170                 }
171         }
172         put_bh(bh);
173         return retval;
174
175 }
176
177 static int update_tind_extent_range(handle_t *handle, struct inode *inode,
178                                     ext4_fsblk_t pblock,
179                                     struct migrate_struct *lb)
180 {
181         struct buffer_head *bh;
182         __le32 *i_data;
183         int i, retval = 0;
184         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
185
186         bh = sb_bread(inode->i_sb, pblock);
187         if (!bh)
188                 return -EIO;
189
190         i_data = (__le32 *)bh->b_data;
191         for (i = 0; i < max_entries; i++) {
192                 if (i_data[i]) {
193                         retval = update_dind_extent_range(handle, inode,
194                                                 le32_to_cpu(i_data[i]), lb);
195                         if (retval)
196                                 break;
197                 } else {
198                         /* Only update the file block number */
199                         lb->curr_block += max_entries * max_entries;
200                 }
201         }
202         put_bh(bh);
203         return retval;
204
205 }
206
207 static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
208 {
209         int retval = 0, needed;
210
211         if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
212                 return 0;
213         /*
214          * We are freeing a blocks. During this we touch
215          * superblock, group descriptor and block bitmap.
216          * So allocate a credit of 3. We may update
217          * quota (user and group).
218          */
219         needed = 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
220
221         if (ext4_journal_extend(handle, needed) != 0)
222                 retval = ext4_journal_restart(handle, needed);
223
224         return retval;
225 }
226
227 static int free_dind_blocks(handle_t *handle,
228                                 struct inode *inode, __le32 i_data)
229 {
230         int i;
231         __le32 *tmp_idata;
232         struct buffer_head *bh;
233         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
234
235         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
236         if (!bh)
237                 return -EIO;
238
239         tmp_idata = (__le32 *)bh->b_data;
240         for (i = 0; i < max_entries; i++) {
241                 if (tmp_idata[i]) {
242                         extend_credit_for_blkdel(handle, inode);
243                         ext4_free_blocks(handle, inode, NULL,
244                                          le32_to_cpu(tmp_idata[i]), 1,
245                                          EXT4_FREE_BLOCKS_METADATA |
246                                          EXT4_FREE_BLOCKS_FORGET);
247                 }
248         }
249         put_bh(bh);
250         extend_credit_for_blkdel(handle, inode);
251         ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
252                          EXT4_FREE_BLOCKS_METADATA |
253                          EXT4_FREE_BLOCKS_FORGET);
254         return 0;
255 }
256
257 static int free_tind_blocks(handle_t *handle,
258                                 struct inode *inode, __le32 i_data)
259 {
260         int i, retval = 0;
261         __le32 *tmp_idata;
262         struct buffer_head *bh;
263         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
264
265         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
266         if (!bh)
267                 return -EIO;
268
269         tmp_idata = (__le32 *)bh->b_data;
270         for (i = 0; i < max_entries; i++) {
271                 if (tmp_idata[i]) {
272                         retval = free_dind_blocks(handle,
273                                         inode, tmp_idata[i]);
274                         if (retval) {
275                                 put_bh(bh);
276                                 return retval;
277                         }
278                 }
279         }
280         put_bh(bh);
281         extend_credit_for_blkdel(handle, inode);
282         ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
283                          EXT4_FREE_BLOCKS_METADATA |
284                          EXT4_FREE_BLOCKS_FORGET);
285         return 0;
286 }
287
288 static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
289 {
290         int retval;
291
292         /* ei->i_data[EXT4_IND_BLOCK] */
293         if (i_data[0]) {
294                 extend_credit_for_blkdel(handle, inode);
295                 ext4_free_blocks(handle, inode, NULL,
296                                 le32_to_cpu(i_data[0]), 1,
297                                  EXT4_FREE_BLOCKS_METADATA |
298                                  EXT4_FREE_BLOCKS_FORGET);
299         }
300
301         /* ei->i_data[EXT4_DIND_BLOCK] */
302         if (i_data[1]) {
303                 retval = free_dind_blocks(handle, inode, i_data[1]);
304                 if (retval)
305                         return retval;
306         }
307
308         /* ei->i_data[EXT4_TIND_BLOCK] */
309         if (i_data[2]) {
310                 retval = free_tind_blocks(handle, inode, i_data[2]);
311                 if (retval)
312                         return retval;
313         }
314         return 0;
315 }
316
317 static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
318                                                 struct inode *tmp_inode)
319 {
320         int retval;
321         __le32  i_data[3];
322         struct ext4_inode_info *ei = EXT4_I(inode);
323         struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
324
325         /*
326          * One credit accounted for writing the
327          * i_data field of the original inode
328          */
329         retval = ext4_journal_extend(handle, 1);
330         if (retval) {
331                 retval = ext4_journal_restart(handle, 1);
332                 if (retval)
333                         goto err_out;
334         }
335
336         i_data[0] = ei->i_data[EXT4_IND_BLOCK];
337         i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
338         i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
339
340         down_write(&EXT4_I(inode)->i_data_sem);
341         /*
342          * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
343          * happened after we started the migrate. We need to
344          * fail the migrate
345          */
346         if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) {
347                 retval = -EAGAIN;
348                 up_write(&EXT4_I(inode)->i_data_sem);
349                 goto err_out;
350         } else
351                 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
352         /*
353          * We have the extent map build with the tmp inode.
354          * Now copy the i_data across
355          */
356         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
357         memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
358
359         /*
360          * Update i_blocks with the new blocks that got
361          * allocated while adding extents for extent index
362          * blocks.
363          *
364          * While converting to extents we need not
365          * update the orignal inode i_blocks for extent blocks
366          * via quota APIs. The quota update happened via tmp_inode already.
367          */
368         spin_lock(&inode->i_lock);
369         inode->i_blocks += tmp_inode->i_blocks;
370         spin_unlock(&inode->i_lock);
371         up_write(&EXT4_I(inode)->i_data_sem);
372
373         /*
374          * We mark the inode dirty after, because we decrement the
375          * i_blocks when freeing the indirect meta-data blocks
376          */
377         retval = free_ind_block(handle, inode, i_data);
378         ext4_mark_inode_dirty(handle, inode);
379
380 err_out:
381         return retval;
382 }
383
384 static int free_ext_idx(handle_t *handle, struct inode *inode,
385                                         struct ext4_extent_idx *ix)
386 {
387         int i, retval = 0;
388         ext4_fsblk_t block;
389         struct buffer_head *bh;
390         struct ext4_extent_header *eh;
391
392         block = ext4_idx_pblock(ix);
393         bh = sb_bread(inode->i_sb, block);
394         if (!bh)
395                 return -EIO;
396
397         eh = (struct ext4_extent_header *)bh->b_data;
398         if (eh->eh_depth != 0) {
399                 ix = EXT_FIRST_INDEX(eh);
400                 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
401                         retval = free_ext_idx(handle, inode, ix);
402                         if (retval)
403                                 break;
404                 }
405         }
406         put_bh(bh);
407         extend_credit_for_blkdel(handle, inode);
408         ext4_free_blocks(handle, inode, NULL, block, 1,
409                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
410         return retval;
411 }
412
413 /*
414  * Free the extent meta data blocks only
415  */
416 static int free_ext_block(handle_t *handle, struct inode *inode)
417 {
418         int i, retval = 0;
419         struct ext4_inode_info *ei = EXT4_I(inode);
420         struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
421         struct ext4_extent_idx *ix;
422         if (eh->eh_depth == 0)
423                 /*
424                  * No extra blocks allocated for extent meta data
425                  */
426                 return 0;
427         ix = EXT_FIRST_INDEX(eh);
428         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
429                 retval = free_ext_idx(handle, inode, ix);
430                 if (retval)
431                         return retval;
432         }
433         return retval;
434 }
435
436 int ext4_ext_migrate(struct inode *inode)
437 {
438         handle_t *handle;
439         int retval = 0, i;
440         __le32 *i_data;
441         struct ext4_inode_info *ei;
442         struct inode *tmp_inode = NULL;
443         struct migrate_struct lb;
444         unsigned long max_entries;
445         __u32 goal;
446         uid_t owner[2];
447
448         /*
449          * If the filesystem does not support extents, or the inode
450          * already is extent-based, error out.
451          */
452         if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
453                                        EXT4_FEATURE_INCOMPAT_EXTENTS) ||
454             (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
455                 return -EINVAL;
456
457         if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
458                 /*
459                  * don't migrate fast symlink
460                  */
461                 return retval;
462
463         /*
464          * Worst case we can touch the allocation bitmaps, a bgd
465          * block, and a block to link in the orphan list.  We do need
466          * need to worry about credits for modifying the quota inode.
467          */
468         handle = ext4_journal_start(inode, EXT4_HT_MIGRATE,
469                 4 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
470
471         if (IS_ERR(handle)) {
472                 retval = PTR_ERR(handle);
473                 return retval;
474         }
475         goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *
476                 EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;
477         owner[0] = i_uid_read(inode);
478         owner[1] = i_gid_read(inode);
479         tmp_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
480                                    S_IFREG, NULL, goal, owner);
481         if (IS_ERR(tmp_inode)) {
482                 retval = PTR_ERR(tmp_inode);
483                 ext4_journal_stop(handle);
484                 return retval;
485         }
486         i_size_write(tmp_inode, i_size_read(inode));
487         /*
488          * Set the i_nlink to zero so it will be deleted later
489          * when we drop inode reference.
490          */
491         clear_nlink(tmp_inode);
492
493         ext4_ext_tree_init(handle, tmp_inode);
494         ext4_orphan_add(handle, tmp_inode);
495         ext4_journal_stop(handle);
496
497         /*
498          * start with one credit accounted for
499          * superblock modification.
500          *
501          * For the tmp_inode we already have committed the
502          * transaction that created the inode. Later as and
503          * when we add extents we extent the journal
504          */
505         /*
506          * Even though we take i_mutex we can still cause block
507          * allocation via mmap write to holes. If we have allocated
508          * new blocks we fail migrate.  New block allocation will
509          * clear EXT4_STATE_EXT_MIGRATE flag.  The flag is updated
510          * with i_data_sem held to prevent racing with block
511          * allocation.
512          */
513         down_read(&EXT4_I(inode)->i_data_sem);
514         ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
515         up_read((&EXT4_I(inode)->i_data_sem));
516
517         handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
518         if (IS_ERR(handle)) {
519                 /*
520                  * It is impossible to update on-disk structures without
521                  * a handle, so just rollback in-core changes and live other
522                  * work to orphan_list_cleanup()
523                  */
524                 ext4_orphan_del(NULL, tmp_inode);
525                 retval = PTR_ERR(handle);
526                 goto out;
527         }
528
529         ei = EXT4_I(inode);
530         i_data = ei->i_data;
531         memset(&lb, 0, sizeof(lb));
532
533         /* 32 bit block address 4 bytes */
534         max_entries = inode->i_sb->s_blocksize >> 2;
535         for (i = 0; i < EXT4_NDIR_BLOCKS; i++) {
536                 if (i_data[i]) {
537                         retval = update_extent_range(handle, tmp_inode,
538                                                 le32_to_cpu(i_data[i]), &lb);
539                         if (retval)
540                                 goto err_out;
541                 } else
542                         lb.curr_block++;
543         }
544         if (i_data[EXT4_IND_BLOCK]) {
545                 retval = update_ind_extent_range(handle, tmp_inode,
546                                 le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb);
547                         if (retval)
548                                 goto err_out;
549         } else
550                 lb.curr_block += max_entries;
551         if (i_data[EXT4_DIND_BLOCK]) {
552                 retval = update_dind_extent_range(handle, tmp_inode,
553                                 le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb);
554                         if (retval)
555                                 goto err_out;
556         } else
557                 lb.curr_block += max_entries * max_entries;
558         if (i_data[EXT4_TIND_BLOCK]) {
559                 retval = update_tind_extent_range(handle, tmp_inode,
560                                 le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb);
561                         if (retval)
562                                 goto err_out;
563         }
564         /*
565          * Build the last extent
566          */
567         retval = finish_range(handle, tmp_inode, &lb);
568 err_out:
569         if (retval)
570                 /*
571                  * Failure case delete the extent information with the
572                  * tmp_inode
573                  */
574                 free_ext_block(handle, tmp_inode);
575         else {
576                 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
577                 if (retval)
578                         /*
579                          * if we fail to swap inode data free the extent
580                          * details of the tmp inode
581                          */
582                         free_ext_block(handle, tmp_inode);
583         }
584
585         /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
586         if (ext4_journal_extend(handle, 1) != 0)
587                 ext4_journal_restart(handle, 1);
588
589         /*
590          * Mark the tmp_inode as of size zero
591          */
592         i_size_write(tmp_inode, 0);
593
594         /*
595          * set the  i_blocks count to zero
596          * so that the ext4_delete_inode does the
597          * right job
598          *
599          * We don't need to take the i_lock because
600          * the inode is not visible to user space.
601          */
602         tmp_inode->i_blocks = 0;
603
604         /* Reset the extent details */
605         ext4_ext_tree_init(handle, tmp_inode);
606         ext4_journal_stop(handle);
607 out:
608         unlock_new_inode(tmp_inode);
609         iput(tmp_inode);
610
611         return retval;
612 }
613
614 /*
615  * Migrate a simple extent-based inode to use the i_blocks[] array
616  */
617 int ext4_ind_migrate(struct inode *inode)
618 {
619         struct ext4_extent_header       *eh;
620         struct ext4_super_block         *es = EXT4_SB(inode->i_sb)->s_es;
621         struct ext4_inode_info          *ei = EXT4_I(inode);
622         struct ext4_extent              *ex;
623         unsigned int                    i, len;
624         ext4_fsblk_t                    blk;
625         handle_t                        *handle;
626         int                             ret;
627
628         if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
629                                        EXT4_FEATURE_INCOMPAT_EXTENTS) ||
630             (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
631                 return -EINVAL;
632
633         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
634                                        EXT4_FEATURE_RO_COMPAT_BIGALLOC))
635                 return -EOPNOTSUPP;
636
637         handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
638         if (IS_ERR(handle))
639                 return PTR_ERR(handle);
640
641         down_write(&EXT4_I(inode)->i_data_sem);
642         ret = ext4_ext_check_inode(inode);
643         if (ret)
644                 goto errout;
645
646         eh = ext_inode_hdr(inode);
647         ex  = EXT_FIRST_EXTENT(eh);
648         if (ext4_blocks_count(es) > EXT4_MAX_BLOCK_FILE_PHYS ||
649             eh->eh_depth != 0 || le16_to_cpu(eh->eh_entries) > 1) {
650                 ret = -EOPNOTSUPP;
651                 goto errout;
652         }
653         if (eh->eh_entries == 0)
654                 blk = len = 0;
655         else {
656                 len = le16_to_cpu(ex->ee_len);
657                 blk = ext4_ext_pblock(ex);
658                 if (len > 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=0; i < len; i++)
667                 ei->i_data[i] = cpu_to_le32(blk++);
668         ext4_mark_inode_dirty(handle, inode);
669 errout:
670         ext4_journal_stop(handle);
671         up_write(&EXT4_I(inode)->i_data_sem);
672         return ret;
673 }