Merge tag 'trace-v5.2-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[sfrench/cifs-2.6.git] / fs / ext4 / migrate.c
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  */
16 struct migrate_struct {
17         ext4_lblk_t first_block, last_block, curr_block;
18         ext4_fsblk_t first_pblock, last_pblock;
19 };
20
21 static 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);
77 err_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
85 static 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
110 static 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
139 static 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
169 static 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
199 static 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
219 static 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
249 static 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
280 static 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
309 static 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
372 err_out:
373         return retval;
374 }
375
376 static 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  */
408 static 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
428 int 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);
559 err_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);
598 out:
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  */
608 int 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);
669 errout:
670         ext4_journal_stop(handle);
671         up_write(&EXT4_I(inode)->i_data_sem);
672         return ret;
673 }