673 lines · c
1// SPDX-License-Identifier: LGPL-2.12/*3 * Copyright IBM Corporation, 20074 * 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 be14 * represented by a single extent15 */16struct migrate_struct {17 ext4_lblk_t first_block, last_block, curr_block;18 ext4_fsblk_t first_pblock, last_pblock;19};20 21static int finish_range(handle_t *handle, struct inode *inode,22 struct migrate_struct *lb)23 24{25 int retval = 0, needed;26 struct ext4_extent newext;27 struct ext4_ext_path *path;28 if (lb->first_pblock == 0)29 return 0;30 31 /* Add the extent to temp inode*/32 newext.ee_block = cpu_to_le32(lb->first_block);33 newext.ee_len = cpu_to_le16(lb->last_block - lb->first_block + 1);34 ext4_ext_store_pblock(&newext, lb->first_pblock);35 /* Locking only for convenience since we are operating on temp inode */36 down_write(&EXT4_I(inode)->i_data_sem);37 path = ext4_find_extent(inode, lb->first_block, NULL, 0);38 if (IS_ERR(path)) {39 retval = PTR_ERR(path);40 goto err_out;41 }42 43 /*44 * Calculate the credit needed to inserting this extent45 * Since we are doing this in loop we may accumulate extra46 * credit. But below we try to not accumulate too much47 * of them by restarting the journal.48 */49 needed = ext4_ext_calc_credits_for_single_extent(inode,50 lb->last_block - lb->first_block + 1, path);51 52 retval = ext4_datasem_ensure_credits(handle, inode, needed, needed, 0);53 if (retval < 0)54 goto err_out;55 path = ext4_ext_insert_extent(handle, inode, path, &newext, 0);56 if (IS_ERR(path))57 retval = PTR_ERR(path);58err_out:59 up_write((&EXT4_I(inode)->i_data_sem));60 ext4_free_ext_path(path);61 lb->first_pblock = 0;62 return retval;63}64 65static int update_extent_range(handle_t *handle, struct inode *inode,66 ext4_fsblk_t pblock, struct migrate_struct *lb)67{68 int retval;69 /*70 * See if we can add on to the existing range (if it exists)71 */72 if (lb->first_pblock &&73 (lb->last_pblock+1 == pblock) &&74 (lb->last_block+1 == lb->curr_block)) {75 lb->last_pblock = pblock;76 lb->last_block = lb->curr_block;77 lb->curr_block++;78 return 0;79 }80 /*81 * Start a new range.82 */83 retval = finish_range(handle, inode, lb);84 lb->first_pblock = lb->last_pblock = pblock;85 lb->first_block = lb->last_block = lb->curr_block;86 lb->curr_block++;87 return retval;88}89 90static int update_ind_extent_range(handle_t *handle, struct inode *inode,91 ext4_fsblk_t pblock,92 struct migrate_struct *lb)93{94 struct buffer_head *bh;95 __le32 *i_data;96 int i, retval = 0;97 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;98 99 bh = ext4_sb_bread(inode->i_sb, pblock, 0);100 if (IS_ERR(bh))101 return PTR_ERR(bh);102 103 i_data = (__le32 *)bh->b_data;104 for (i = 0; i < max_entries; i++) {105 if (i_data[i]) {106 retval = update_extent_range(handle, inode,107 le32_to_cpu(i_data[i]), lb);108 if (retval)109 break;110 } else {111 lb->curr_block++;112 }113 }114 put_bh(bh);115 return retval;116 117}118 119static int update_dind_extent_range(handle_t *handle, struct inode *inode,120 ext4_fsblk_t pblock,121 struct migrate_struct *lb)122{123 struct buffer_head *bh;124 __le32 *i_data;125 int i, retval = 0;126 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;127 128 bh = ext4_sb_bread(inode->i_sb, pblock, 0);129 if (IS_ERR(bh))130 return PTR_ERR(bh);131 132 i_data = (__le32 *)bh->b_data;133 for (i = 0; i < max_entries; i++) {134 if (i_data[i]) {135 retval = update_ind_extent_range(handle, inode,136 le32_to_cpu(i_data[i]), lb);137 if (retval)138 break;139 } else {140 /* Only update the file block number */141 lb->curr_block += max_entries;142 }143 }144 put_bh(bh);145 return retval;146 147}148 149static int update_tind_extent_range(handle_t *handle, struct inode *inode,150 ext4_fsblk_t pblock,151 struct migrate_struct *lb)152{153 struct buffer_head *bh;154 __le32 *i_data;155 int i, retval = 0;156 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;157 158 bh = ext4_sb_bread(inode->i_sb, pblock, 0);159 if (IS_ERR(bh))160 return PTR_ERR(bh);161 162 i_data = (__le32 *)bh->b_data;163 for (i = 0; i < max_entries; i++) {164 if (i_data[i]) {165 retval = update_dind_extent_range(handle, inode,166 le32_to_cpu(i_data[i]), lb);167 if (retval)168 break;169 } else {170 /* Only update the file block number */171 lb->curr_block += max_entries * max_entries;172 }173 }174 put_bh(bh);175 return retval;176 177}178 179static int free_dind_blocks(handle_t *handle,180 struct inode *inode, __le32 i_data)181{182 int i;183 __le32 *tmp_idata;184 struct buffer_head *bh;185 struct super_block *sb = inode->i_sb;186 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;187 int err;188 189 bh = ext4_sb_bread(sb, le32_to_cpu(i_data), 0);190 if (IS_ERR(bh))191 return PTR_ERR(bh);192 193 tmp_idata = (__le32 *)bh->b_data;194 for (i = 0; i < max_entries; i++) {195 if (tmp_idata[i]) {196 err = ext4_journal_ensure_credits(handle,197 EXT4_RESERVE_TRANS_BLOCKS,198 ext4_free_metadata_revoke_credits(sb, 1));199 if (err < 0) {200 put_bh(bh);201 return err;202 }203 ext4_free_blocks(handle, inode, NULL,204 le32_to_cpu(tmp_idata[i]), 1,205 EXT4_FREE_BLOCKS_METADATA |206 EXT4_FREE_BLOCKS_FORGET);207 }208 }209 put_bh(bh);210 err = ext4_journal_ensure_credits(handle, EXT4_RESERVE_TRANS_BLOCKS,211 ext4_free_metadata_revoke_credits(sb, 1));212 if (err < 0)213 return err;214 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,215 EXT4_FREE_BLOCKS_METADATA |216 EXT4_FREE_BLOCKS_FORGET);217 return 0;218}219 220static int free_tind_blocks(handle_t *handle,221 struct inode *inode, __le32 i_data)222{223 int i, retval = 0;224 __le32 *tmp_idata;225 struct buffer_head *bh;226 unsigned long max_entries = inode->i_sb->s_blocksize >> 2;227 228 bh = ext4_sb_bread(inode->i_sb, le32_to_cpu(i_data), 0);229 if (IS_ERR(bh))230 return PTR_ERR(bh);231 232 tmp_idata = (__le32 *)bh->b_data;233 for (i = 0; i < max_entries; i++) {234 if (tmp_idata[i]) {235 retval = free_dind_blocks(handle,236 inode, tmp_idata[i]);237 if (retval) {238 put_bh(bh);239 return retval;240 }241 }242 }243 put_bh(bh);244 retval = ext4_journal_ensure_credits(handle, EXT4_RESERVE_TRANS_BLOCKS,245 ext4_free_metadata_revoke_credits(inode->i_sb, 1));246 if (retval < 0)247 return retval;248 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,249 EXT4_FREE_BLOCKS_METADATA |250 EXT4_FREE_BLOCKS_FORGET);251 return 0;252}253 254static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)255{256 int retval;257 258 /* ei->i_data[EXT4_IND_BLOCK] */259 if (i_data[0]) {260 retval = ext4_journal_ensure_credits(handle,261 EXT4_RESERVE_TRANS_BLOCKS,262 ext4_free_metadata_revoke_credits(inode->i_sb, 1));263 if (retval < 0)264 return retval;265 ext4_free_blocks(handle, inode, NULL,266 le32_to_cpu(i_data[0]), 1,267 EXT4_FREE_BLOCKS_METADATA |268 EXT4_FREE_BLOCKS_FORGET);269 }270 271 /* ei->i_data[EXT4_DIND_BLOCK] */272 if (i_data[1]) {273 retval = free_dind_blocks(handle, inode, i_data[1]);274 if (retval)275 return retval;276 }277 278 /* ei->i_data[EXT4_TIND_BLOCK] */279 if (i_data[2]) {280 retval = free_tind_blocks(handle, inode, i_data[2]);281 if (retval)282 return retval;283 }284 return 0;285}286 287static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,288 struct inode *tmp_inode)289{290 int retval, retval2 = 0;291 __le32 i_data[3];292 struct ext4_inode_info *ei = EXT4_I(inode);293 struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);294 295 /*296 * One credit accounted for writing the297 * i_data field of the original inode298 */299 retval = ext4_journal_ensure_credits(handle, 1, 0);300 if (retval < 0)301 goto err_out;302 303 i_data[0] = ei->i_data[EXT4_IND_BLOCK];304 i_data[1] = ei->i_data[EXT4_DIND_BLOCK];305 i_data[2] = ei->i_data[EXT4_TIND_BLOCK];306 307 down_write(&EXT4_I(inode)->i_data_sem);308 /*309 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation310 * happened after we started the migrate. We need to311 * fail the migrate312 */313 if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) {314 retval = -EAGAIN;315 up_write(&EXT4_I(inode)->i_data_sem);316 goto err_out;317 } else318 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);319 /*320 * We have the extent map build with the tmp inode.321 * Now copy the i_data across322 */323 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);324 memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));325 326 /*327 * Update i_blocks with the new blocks that got328 * allocated while adding extents for extent index329 * blocks.330 *331 * While converting to extents we need not332 * update the original inode i_blocks for extent blocks333 * via quota APIs. The quota update happened via tmp_inode already.334 */335 spin_lock(&inode->i_lock);336 inode->i_blocks += tmp_inode->i_blocks;337 spin_unlock(&inode->i_lock);338 up_write(&EXT4_I(inode)->i_data_sem);339 340 /*341 * We mark the inode dirty after, because we decrement the342 * i_blocks when freeing the indirect meta-data blocks343 */344 retval = free_ind_block(handle, inode, i_data);345 retval2 = ext4_mark_inode_dirty(handle, inode);346 if (unlikely(retval2 && !retval))347 retval = retval2;348 349err_out:350 return retval;351}352 353static int free_ext_idx(handle_t *handle, struct inode *inode,354 struct ext4_extent_idx *ix)355{356 int i, retval = 0;357 ext4_fsblk_t block;358 struct buffer_head *bh;359 struct ext4_extent_header *eh;360 361 block = ext4_idx_pblock(ix);362 bh = ext4_sb_bread(inode->i_sb, block, 0);363 if (IS_ERR(bh))364 return PTR_ERR(bh);365 366 eh = (struct ext4_extent_header *)bh->b_data;367 if (eh->eh_depth != 0) {368 ix = EXT_FIRST_INDEX(eh);369 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {370 retval = free_ext_idx(handle, inode, ix);371 if (retval) {372 put_bh(bh);373 return retval;374 }375 }376 }377 put_bh(bh);378 retval = ext4_journal_ensure_credits(handle, EXT4_RESERVE_TRANS_BLOCKS,379 ext4_free_metadata_revoke_credits(inode->i_sb, 1));380 if (retval < 0)381 return retval;382 ext4_free_blocks(handle, inode, NULL, block, 1,383 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);384 return 0;385}386 387/*388 * Free the extent meta data blocks only389 */390static int free_ext_block(handle_t *handle, struct inode *inode)391{392 int i, retval = 0;393 struct ext4_inode_info *ei = EXT4_I(inode);394 struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;395 struct ext4_extent_idx *ix;396 if (eh->eh_depth == 0)397 /*398 * No extra blocks allocated for extent meta data399 */400 return 0;401 ix = EXT_FIRST_INDEX(eh);402 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {403 retval = free_ext_idx(handle, inode, ix);404 if (retval)405 return retval;406 }407 return retval;408}409 410int ext4_ext_migrate(struct inode *inode)411{412 handle_t *handle;413 int retval = 0, i;414 __le32 *i_data;415 struct ext4_inode_info *ei;416 struct inode *tmp_inode = NULL;417 struct migrate_struct lb;418 unsigned long max_entries;419 __u32 goal, tmp_csum_seed;420 uid_t owner[2];421 int alloc_ctx;422 423 /*424 * If the filesystem does not support extents, or the inode425 * already is extent-based, error out.426 */427 if (!ext4_has_feature_extents(inode->i_sb) ||428 ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||429 ext4_has_inline_data(inode))430 return -EINVAL;431 432 if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)433 /*434 * don't migrate fast symlink435 */436 return retval;437 438 alloc_ctx = ext4_writepages_down_write(inode->i_sb);439 440 /*441 * Worst case we can touch the allocation bitmaps and a block442 * group descriptor block. We do need to worry about443 * credits for modifying the quota inode.444 */445 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE,446 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));447 448 if (IS_ERR(handle)) {449 retval = PTR_ERR(handle);450 goto out_unlock;451 }452 goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *453 EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;454 owner[0] = i_uid_read(inode);455 owner[1] = i_gid_read(inode);456 tmp_inode = ext4_new_inode(handle, d_inode(inode->i_sb->s_root),457 S_IFREG, NULL, goal, owner, 0);458 if (IS_ERR(tmp_inode)) {459 retval = PTR_ERR(tmp_inode);460 ext4_journal_stop(handle);461 goto out_unlock;462 }463 /*464 * Use the correct seed for checksum (i.e. the seed from 'inode'). This465 * is so that the metadata blocks will have the correct checksum after466 * the migration.467 */468 ei = EXT4_I(inode);469 tmp_csum_seed = EXT4_I(tmp_inode)->i_csum_seed;470 EXT4_I(tmp_inode)->i_csum_seed = ei->i_csum_seed;471 i_size_write(tmp_inode, i_size_read(inode));472 /*473 * Set the i_nlink to zero so it will be deleted later474 * when we drop inode reference.475 */476 clear_nlink(tmp_inode);477 478 ext4_ext_tree_init(handle, tmp_inode);479 ext4_journal_stop(handle);480 481 /*482 * start with one credit accounted for483 * superblock modification.484 *485 * For the tmp_inode we already have committed the486 * transaction that created the inode. Later as and487 * when we add extents we extent the journal488 */489 /*490 * Even though we take i_rwsem we can still cause block491 * allocation via mmap write to holes. If we have allocated492 * new blocks we fail migrate. New block allocation will493 * clear EXT4_STATE_EXT_MIGRATE flag. The flag is updated494 * with i_data_sem held to prevent racing with block495 * allocation.496 */497 down_read(&EXT4_I(inode)->i_data_sem);498 ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE);499 up_read((&EXT4_I(inode)->i_data_sem));500 501 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);502 if (IS_ERR(handle)) {503 retval = PTR_ERR(handle);504 goto out_tmp_inode;505 }506 507 i_data = ei->i_data;508 memset(&lb, 0, sizeof(lb));509 510 /* 32 bit block address 4 bytes */511 max_entries = inode->i_sb->s_blocksize >> 2;512 for (i = 0; i < EXT4_NDIR_BLOCKS; i++) {513 if (i_data[i]) {514 retval = update_extent_range(handle, tmp_inode,515 le32_to_cpu(i_data[i]), &lb);516 if (retval)517 goto err_out;518 } else519 lb.curr_block++;520 }521 if (i_data[EXT4_IND_BLOCK]) {522 retval = update_ind_extent_range(handle, tmp_inode,523 le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb);524 if (retval)525 goto err_out;526 } else527 lb.curr_block += max_entries;528 if (i_data[EXT4_DIND_BLOCK]) {529 retval = update_dind_extent_range(handle, tmp_inode,530 le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb);531 if (retval)532 goto err_out;533 } else534 lb.curr_block += max_entries * max_entries;535 if (i_data[EXT4_TIND_BLOCK]) {536 retval = update_tind_extent_range(handle, tmp_inode,537 le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb);538 if (retval)539 goto err_out;540 }541 /*542 * Build the last extent543 */544 retval = finish_range(handle, tmp_inode, &lb);545err_out:546 if (retval)547 /*548 * Failure case delete the extent information with the549 * tmp_inode550 */551 free_ext_block(handle, tmp_inode);552 else {553 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);554 if (retval)555 /*556 * if we fail to swap inode data free the extent557 * details of the tmp inode558 */559 free_ext_block(handle, tmp_inode);560 }561 562 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */563 retval = ext4_journal_ensure_credits(handle, 1, 0);564 if (retval < 0)565 goto out_stop;566 /*567 * Mark the tmp_inode as of size zero568 */569 i_size_write(tmp_inode, 0);570 571 /*572 * set the i_blocks count to zero573 * so that the ext4_evict_inode() does the574 * right job575 *576 * We don't need to take the i_lock because577 * the inode is not visible to user space.578 */579 tmp_inode->i_blocks = 0;580 EXT4_I(tmp_inode)->i_csum_seed = tmp_csum_seed;581 582 /* Reset the extent details */583 ext4_ext_tree_init(handle, tmp_inode);584out_stop:585 ext4_journal_stop(handle);586out_tmp_inode:587 unlock_new_inode(tmp_inode);588 iput(tmp_inode);589out_unlock:590 ext4_writepages_up_write(inode->i_sb, alloc_ctx);591 return retval;592}593 594/*595 * Migrate a simple extent-based inode to use the i_blocks[] array596 */597int ext4_ind_migrate(struct inode *inode)598{599 struct ext4_extent_header *eh;600 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);601 struct ext4_super_block *es = sbi->s_es;602 struct ext4_inode_info *ei = EXT4_I(inode);603 struct ext4_extent *ex;604 unsigned int i, len;605 ext4_lblk_t start, end;606 ext4_fsblk_t blk;607 handle_t *handle;608 int ret, ret2 = 0;609 int alloc_ctx;610 611 if (!ext4_has_feature_extents(inode->i_sb) ||612 (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))613 return -EINVAL;614 615 if (ext4_has_feature_bigalloc(inode->i_sb))616 return -EOPNOTSUPP;617 618 /*619 * In order to get correct extent info, force all delayed allocation620 * blocks to be allocated, otherwise delayed allocation blocks may not621 * be reflected and bypass the checks on extent header.622 */623 if (test_opt(inode->i_sb, DELALLOC))624 ext4_alloc_da_blocks(inode);625 626 alloc_ctx = ext4_writepages_down_write(inode->i_sb);627 628 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);629 if (IS_ERR(handle)) {630 ret = PTR_ERR(handle);631 goto out_unlock;632 }633 634 down_write(&EXT4_I(inode)->i_data_sem);635 ret = ext4_ext_check_inode(inode);636 if (ret)637 goto errout;638 639 eh = ext_inode_hdr(inode);640 ex = EXT_FIRST_EXTENT(eh);641 if (ext4_blocks_count(es) > EXT4_MAX_BLOCK_FILE_PHYS ||642 eh->eh_depth != 0 || le16_to_cpu(eh->eh_entries) > 1) {643 ret = -EOPNOTSUPP;644 goto errout;645 }646 if (eh->eh_entries == 0)647 blk = len = start = end = 0;648 else {649 len = le16_to_cpu(ex->ee_len);650 blk = ext4_ext_pblock(ex);651 start = le32_to_cpu(ex->ee_block);652 end = start + len - 1;653 if (end >= 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 = start; i <= end; i++)662 ei->i_data[i] = cpu_to_le32(blk++);663 ret2 = ext4_mark_inode_dirty(handle, inode);664 if (unlikely(ret2 && !ret))665 ret = ret2;666errout:667 up_write(&EXT4_I(inode)->i_data_sem);668 ext4_journal_stop(handle);669out_unlock:670 ext4_writepages_up_write(inode->i_sb, alloc_ctx);671 return ret;672}673