mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / ext4 / migrate.c
blob0d785868cc503e82042294eec95462f673143890
1 /*
2 * Copyright IBM Corporation, 2007
3 * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
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.
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.
15 #include <linux/slab.h>
16 #include "ext4_jbd2.h"
17 #include "ext4_extents.h"
20 * The contiguous blocks details which can be
21 * represented by a single extent
23 struct migrate_struct {
24 ext4_lblk_t first_block, last_block, curr_block;
25 ext4_fsblk_t first_pblock, last_pblock;
28 static int finish_range(handle_t *handle, struct inode *inode,
29 struct migrate_struct *lb)
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;
38 /* Add the extent to temp inode*/
39 newext.ee_block = cpu_to_le32(lb->first_block);
40 newext.ee_len = cpu_to_le16(lb->last_block - lb->first_block + 1);
41 ext4_ext_store_pblock(&newext, lb->first_pblock);
42 /* Locking only for convinience since we are operating on temp inode */
43 down_write(&EXT4_I(inode)->i_data_sem);
44 path = ext4_find_extent(inode, lb->first_block, NULL, 0);
45 if (IS_ERR(path)) {
46 retval = PTR_ERR(path);
47 path = NULL;
48 goto err_out;
52 * Calculate the credit needed to inserting this extent
53 * Since we are doing this in loop we may accumalate extra
54 * credit. But below we try to not accumalate too much
55 * of them by restarting the journal.
57 needed = ext4_ext_calc_credits_for_single_extent(inode,
58 lb->last_block - lb->first_block + 1, path);
61 * Make sure the credit we accumalated is not really high
63 if (needed && ext4_handle_has_enough_credits(handle,
64 EXT4_RESERVE_TRANS_BLOCKS)) {
65 up_write((&EXT4_I(inode)->i_data_sem));
66 retval = ext4_journal_restart(handle, needed);
67 down_write((&EXT4_I(inode)->i_data_sem));
68 if (retval)
69 goto err_out;
70 } else if (needed) {
71 retval = ext4_journal_extend(handle, needed);
72 if (retval) {
74 * IF not able to extend the journal restart the journal
76 up_write((&EXT4_I(inode)->i_data_sem));
77 retval = ext4_journal_restart(handle, needed);
78 down_write((&EXT4_I(inode)->i_data_sem));
79 if (retval)
80 goto err_out;
83 retval = ext4_ext_insert_extent(handle, inode, &path, &newext, 0);
84 err_out:
85 up_write((&EXT4_I(inode)->i_data_sem));
86 ext4_ext_drop_refs(path);
87 kfree(path);
88 lb->first_pblock = 0;
89 return retval;
92 static int update_extent_range(handle_t *handle, struct inode *inode,
93 ext4_fsblk_t pblock, struct migrate_struct *lb)
95 int retval;
97 * See if we can add on to the existing range (if it exists)
99 if (lb->first_pblock &&
100 (lb->last_pblock+1 == pblock) &&
101 (lb->last_block+1 == lb->curr_block)) {
102 lb->last_pblock = pblock;
103 lb->last_block = lb->curr_block;
104 lb->curr_block++;
105 return 0;
108 * Start a new range.
110 retval = finish_range(handle, inode, lb);
111 lb->first_pblock = lb->last_pblock = pblock;
112 lb->first_block = lb->last_block = lb->curr_block;
113 lb->curr_block++;
114 return retval;
117 static int update_ind_extent_range(handle_t *handle, struct inode *inode,
118 ext4_fsblk_t pblock,
119 struct migrate_struct *lb)
121 struct buffer_head *bh;
122 __le32 *i_data;
123 int i, retval = 0;
124 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
126 bh = ext4_sb_bread(inode->i_sb, pblock, 0);
127 if (IS_ERR(bh))
128 return PTR_ERR(bh);
130 i_data = (__le32 *)bh->b_data;
131 for (i = 0; i < max_entries; i++) {
132 if (i_data[i]) {
133 retval = update_extent_range(handle, inode,
134 le32_to_cpu(i_data[i]), lb);
135 if (retval)
136 break;
137 } else {
138 lb->curr_block++;
141 put_bh(bh);
142 return retval;
146 static int update_dind_extent_range(handle_t *handle, struct inode *inode,
147 ext4_fsblk_t pblock,
148 struct migrate_struct *lb)
150 struct buffer_head *bh;
151 __le32 *i_data;
152 int i, retval = 0;
153 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
155 bh = ext4_sb_bread(inode->i_sb, pblock, 0);
156 if (IS_ERR(bh))
157 return PTR_ERR(bh);
159 i_data = (__le32 *)bh->b_data;
160 for (i = 0; i < max_entries; i++) {
161 if (i_data[i]) {
162 retval = update_ind_extent_range(handle, inode,
163 le32_to_cpu(i_data[i]), lb);
164 if (retval)
165 break;
166 } else {
167 /* Only update the file block number */
168 lb->curr_block += max_entries;
171 put_bh(bh);
172 return retval;
176 static int update_tind_extent_range(handle_t *handle, struct inode *inode,
177 ext4_fsblk_t pblock,
178 struct migrate_struct *lb)
180 struct buffer_head *bh;
181 __le32 *i_data;
182 int i, retval = 0;
183 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
185 bh = ext4_sb_bread(inode->i_sb, pblock, 0);
186 if (IS_ERR(bh))
187 return PTR_ERR(bh);
189 i_data = (__le32 *)bh->b_data;
190 for (i = 0; i < max_entries; i++) {
191 if (i_data[i]) {
192 retval = update_dind_extent_range(handle, inode,
193 le32_to_cpu(i_data[i]), lb);
194 if (retval)
195 break;
196 } else {
197 /* Only update the file block number */
198 lb->curr_block += max_entries * max_entries;
201 put_bh(bh);
202 return retval;
206 static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
208 int retval = 0, needed;
210 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
211 return 0;
213 * We are freeing a blocks. During this we touch
214 * superblock, group descriptor and block bitmap.
215 * So allocate a credit of 3. We may update
216 * quota (user and group).
218 needed = 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
220 if (ext4_journal_extend(handle, needed) != 0)
221 retval = ext4_journal_restart(handle, needed);
223 return retval;
226 static int free_dind_blocks(handle_t *handle,
227 struct inode *inode, __le32 i_data)
229 int i;
230 __le32 *tmp_idata;
231 struct buffer_head *bh;
232 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
234 bh = ext4_sb_bread(inode->i_sb, le32_to_cpu(i_data), 0);
235 if (IS_ERR(bh))
236 return PTR_ERR(bh);
238 tmp_idata = (__le32 *)bh->b_data;
239 for (i = 0; i < max_entries; i++) {
240 if (tmp_idata[i]) {
241 extend_credit_for_blkdel(handle, inode);
242 ext4_free_blocks(handle, inode, NULL,
243 le32_to_cpu(tmp_idata[i]), 1,
244 EXT4_FREE_BLOCKS_METADATA |
245 EXT4_FREE_BLOCKS_FORGET);
248 put_bh(bh);
249 extend_credit_for_blkdel(handle, inode);
250 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
251 EXT4_FREE_BLOCKS_METADATA |
252 EXT4_FREE_BLOCKS_FORGET);
253 return 0;
256 static int free_tind_blocks(handle_t *handle,
257 struct inode *inode, __le32 i_data)
259 int i, retval = 0;
260 __le32 *tmp_idata;
261 struct buffer_head *bh;
262 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
264 bh = ext4_sb_bread(inode->i_sb, le32_to_cpu(i_data), 0);
265 if (IS_ERR(bh))
266 return PTR_ERR(bh);
268 tmp_idata = (__le32 *)bh->b_data;
269 for (i = 0; i < max_entries; i++) {
270 if (tmp_idata[i]) {
271 retval = free_dind_blocks(handle,
272 inode, tmp_idata[i]);
273 if (retval) {
274 put_bh(bh);
275 return retval;
279 put_bh(bh);
280 extend_credit_for_blkdel(handle, inode);
281 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
282 EXT4_FREE_BLOCKS_METADATA |
283 EXT4_FREE_BLOCKS_FORGET);
284 return 0;
287 static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
289 int retval;
291 /* ei->i_data[EXT4_IND_BLOCK] */
292 if (i_data[0]) {
293 extend_credit_for_blkdel(handle, inode);
294 ext4_free_blocks(handle, inode, NULL,
295 le32_to_cpu(i_data[0]), 1,
296 EXT4_FREE_BLOCKS_METADATA |
297 EXT4_FREE_BLOCKS_FORGET);
300 /* ei->i_data[EXT4_DIND_BLOCK] */
301 if (i_data[1]) {
302 retval = free_dind_blocks(handle, inode, i_data[1]);
303 if (retval)
304 return retval;
307 /* ei->i_data[EXT4_TIND_BLOCK] */
308 if (i_data[2]) {
309 retval = free_tind_blocks(handle, inode, i_data[2]);
310 if (retval)
311 return retval;
313 return 0;
316 static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
317 struct inode *tmp_inode)
319 int retval;
320 __le32 i_data[3];
321 struct ext4_inode_info *ei = EXT4_I(inode);
322 struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
325 * One credit accounted for writing the
326 * i_data field of the original inode
328 retval = ext4_journal_extend(handle, 1);
329 if (retval) {
330 retval = ext4_journal_restart(handle, 1);
331 if (retval)
332 goto err_out;
335 i_data[0] = ei->i_data[EXT4_IND_BLOCK];
336 i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
337 i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
339 down_write(&EXT4_I(inode)->i_data_sem);
341 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
342 * happened after we started the migrate. We need to
343 * fail the migrate
345 if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) {
346 retval = -EAGAIN;
347 up_write(&EXT4_I(inode)->i_data_sem);
348 goto err_out;
349 } else
350 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
352 * We have the extent map build with the tmp inode.
353 * Now copy the i_data across
355 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
356 memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
359 * Update i_blocks with the new blocks that got
360 * allocated while adding extents for extent index
361 * blocks.
363 * While converting to extents we need not
364 * update the original inode i_blocks for extent blocks
365 * via quota APIs. The quota update happened via tmp_inode already.
367 spin_lock(&inode->i_lock);
368 inode->i_blocks += tmp_inode->i_blocks;
369 spin_unlock(&inode->i_lock);
370 up_write(&EXT4_I(inode)->i_data_sem);
373 * We mark the inode dirty after, because we decrement the
374 * i_blocks when freeing the indirect meta-data blocks
376 retval = free_ind_block(handle, inode, i_data);
377 ext4_mark_inode_dirty(handle, inode);
379 err_out:
380 return retval;
383 static int free_ext_idx(handle_t *handle, struct inode *inode,
384 struct ext4_extent_idx *ix)
386 int i, retval = 0;
387 ext4_fsblk_t block;
388 struct buffer_head *bh;
389 struct ext4_extent_header *eh;
391 block = ext4_idx_pblock(ix);
392 bh = ext4_sb_bread(inode->i_sb, block, 0);
393 if (IS_ERR(bh))
394 return PTR_ERR(bh);
396 eh = (struct ext4_extent_header *)bh->b_data;
397 if (eh->eh_depth != 0) {
398 ix = EXT_FIRST_INDEX(eh);
399 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
400 retval = free_ext_idx(handle, inode, ix);
401 if (retval)
402 break;
405 put_bh(bh);
406 extend_credit_for_blkdel(handle, inode);
407 ext4_free_blocks(handle, inode, NULL, block, 1,
408 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
409 return retval;
413 * Free the extent meta data blocks only
415 static int free_ext_block(handle_t *handle, struct inode *inode)
417 int i, retval = 0;
418 struct ext4_inode_info *ei = EXT4_I(inode);
419 struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
420 struct ext4_extent_idx *ix;
421 if (eh->eh_depth == 0)
423 * No extra blocks allocated for extent meta data
425 return 0;
426 ix = EXT_FIRST_INDEX(eh);
427 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
428 retval = free_ext_idx(handle, inode, ix);
429 if (retval)
430 return retval;
432 return retval;
435 int ext4_ext_migrate(struct inode *inode)
437 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
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];
449 * If the filesystem does not support extents, or the inode
450 * already is extent-based, error out.
452 if (!ext4_has_feature_extents(inode->i_sb) ||
453 (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
454 return -EINVAL;
456 if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
458 * don't migrate fast symlink
460 return retval;
462 percpu_down_write(&sbi->s_writepages_rwsem);
465 * Worst case we can touch the allocation bitmaps, a bgd
466 * block, and a block to link in the orphan list. We do need
467 * need to worry about credits for modifying the quota inode.
469 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE,
470 4 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
472 if (IS_ERR(handle)) {
473 retval = PTR_ERR(handle);
474 goto out_unlock;
476 goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *
477 EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;
478 owner[0] = i_uid_read(inode);
479 owner[1] = i_gid_read(inode);
480 tmp_inode = ext4_new_inode(handle, d_inode(inode->i_sb->s_root),
481 S_IFREG, NULL, goal, owner, 0);
482 if (IS_ERR(tmp_inode)) {
483 retval = PTR_ERR(tmp_inode);
484 ext4_journal_stop(handle);
485 goto out_unlock;
487 i_size_write(tmp_inode, i_size_read(inode));
489 * Set the i_nlink to zero so it will be deleted later
490 * when we drop inode reference.
492 clear_nlink(tmp_inode);
494 ext4_ext_tree_init(handle, tmp_inode);
495 ext4_orphan_add(handle, tmp_inode);
496 ext4_journal_stop(handle);
499 * start with one credit accounted for
500 * superblock modification.
502 * For the tmp_inode we already have committed the
503 * transaction that created the inode. Later as and
504 * when we add extents we extent the journal
507 * Even though we take i_mutex we can still cause block
508 * allocation via mmap write to holes. If we have allocated
509 * new blocks we fail migrate. New block allocation will
510 * clear EXT4_STATE_EXT_MIGRATE flag. The flag is updated
511 * with i_data_sem held to prevent racing with block
512 * allocation.
514 down_read(&EXT4_I(inode)->i_data_sem);
515 ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
516 up_read((&EXT4_I(inode)->i_data_sem));
518 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
519 if (IS_ERR(handle)) {
521 * It is impossible to update on-disk structures without
522 * a handle, so just rollback in-core changes and live other
523 * work to orphan_list_cleanup()
525 ext4_orphan_del(NULL, tmp_inode);
526 retval = PTR_ERR(handle);
527 goto out_tmp_inode;
530 ei = EXT4_I(inode);
531 i_data = ei->i_data;
532 memset(&lb, 0, sizeof(lb));
534 /* 32 bit block address 4 bytes */
535 max_entries = inode->i_sb->s_blocksize >> 2;
536 for (i = 0; i < EXT4_NDIR_BLOCKS; i++) {
537 if (i_data[i]) {
538 retval = update_extent_range(handle, tmp_inode,
539 le32_to_cpu(i_data[i]), &lb);
540 if (retval)
541 goto err_out;
542 } else
543 lb.curr_block++;
545 if (i_data[EXT4_IND_BLOCK]) {
546 retval = update_ind_extent_range(handle, tmp_inode,
547 le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb);
548 if (retval)
549 goto err_out;
550 } else
551 lb.curr_block += max_entries;
552 if (i_data[EXT4_DIND_BLOCK]) {
553 retval = update_dind_extent_range(handle, tmp_inode,
554 le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb);
555 if (retval)
556 goto err_out;
557 } else
558 lb.curr_block += max_entries * max_entries;
559 if (i_data[EXT4_TIND_BLOCK]) {
560 retval = update_tind_extent_range(handle, tmp_inode,
561 le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb);
562 if (retval)
563 goto err_out;
566 * Build the last extent
568 retval = finish_range(handle, tmp_inode, &lb);
569 err_out:
570 if (retval)
572 * Failure case delete the extent information with the
573 * tmp_inode
575 free_ext_block(handle, tmp_inode);
576 else {
577 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
578 if (retval)
580 * if we fail to swap inode data free the extent
581 * details of the tmp inode
583 free_ext_block(handle, tmp_inode);
586 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
587 if (ext4_journal_extend(handle, 1) != 0)
588 ext4_journal_restart(handle, 1);
591 * Mark the tmp_inode as of size zero
593 i_size_write(tmp_inode, 0);
596 * set the i_blocks count to zero
597 * so that the ext4_evict_inode() does the
598 * right job
600 * We don't need to take the i_lock because
601 * the inode is not visible to user space.
603 tmp_inode->i_blocks = 0;
605 /* Reset the extent details */
606 ext4_ext_tree_init(handle, tmp_inode);
607 ext4_journal_stop(handle);
608 out_tmp_inode:
609 unlock_new_inode(tmp_inode);
610 iput(tmp_inode);
611 out_unlock:
612 percpu_up_write(&sbi->s_writepages_rwsem);
613 return retval;
617 * Migrate a simple extent-based inode to use the i_blocks[] array
619 int ext4_ind_migrate(struct inode *inode)
621 struct ext4_extent_header *eh;
622 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
623 struct ext4_super_block *es = sbi->s_es;
624 struct ext4_inode_info *ei = EXT4_I(inode);
625 struct ext4_extent *ex;
626 unsigned int i, len;
627 ext4_lblk_t start, end;
628 ext4_fsblk_t blk;
629 handle_t *handle;
630 int ret;
632 if (!ext4_has_feature_extents(inode->i_sb) ||
633 (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
634 return -EINVAL;
636 if (ext4_has_feature_bigalloc(inode->i_sb))
637 return -EOPNOTSUPP;
640 * In order to get correct extent info, force all delayed allocation
641 * blocks to be allocated, otherwise delayed allocation blocks may not
642 * be reflected and bypass the checks on extent header.
644 if (test_opt(inode->i_sb, DELALLOC))
645 ext4_alloc_da_blocks(inode);
647 percpu_down_write(&sbi->s_writepages_rwsem);
649 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
650 if (IS_ERR(handle)) {
651 ret = PTR_ERR(handle);
652 goto out_unlock;
655 down_write(&EXT4_I(inode)->i_data_sem);
656 ret = ext4_ext_check_inode(inode);
657 if (ret)
658 goto errout;
660 eh = ext_inode_hdr(inode);
661 ex = EXT_FIRST_EXTENT(eh);
662 if (ext4_blocks_count(es) > EXT4_MAX_BLOCK_FILE_PHYS ||
663 eh->eh_depth != 0 || le16_to_cpu(eh->eh_entries) > 1) {
664 ret = -EOPNOTSUPP;
665 goto errout;
667 if (eh->eh_entries == 0)
668 blk = len = start = end = 0;
669 else {
670 len = le16_to_cpu(ex->ee_len);
671 blk = ext4_ext_pblock(ex);
672 start = le32_to_cpu(ex->ee_block);
673 end = start + len - 1;
674 if (end >= EXT4_NDIR_BLOCKS) {
675 ret = -EOPNOTSUPP;
676 goto errout;
680 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
681 memset(ei->i_data, 0, sizeof(ei->i_data));
682 for (i = start; i <= end; i++)
683 ei->i_data[i] = cpu_to_le32(blk++);
684 ext4_mark_inode_dirty(handle, inode);
685 errout:
686 ext4_journal_stop(handle);
687 up_write(&EXT4_I(inode)->i_data_sem);
688 out_unlock:
689 percpu_up_write(&sbi->s_writepages_rwsem);
690 return ret;