1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2011 Oracle. All rights reserved.
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public
10 * License version 2 as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
18 #include <linux/types.h>
19 #include <linux/mount.h>
20 #include <linux/swap.h>
22 #include <cluster/masklog.h>
25 #include "ocfs2_ioctl.h"
30 #include "extent_map.h"
37 #include "buffer_head_io.h"
40 #include "refcounttree.h"
41 #include "move_extents.h"
43 struct ocfs2_move_extents_context {
52 struct ocfs2_move_extents *range;
53 struct ocfs2_extent_tree et;
54 struct ocfs2_alloc_context *meta_ac;
55 struct ocfs2_alloc_context *data_ac;
56 struct ocfs2_cached_dealloc_ctxt dealloc;
59 static int __ocfs2_move_extent(handle_t *handle,
60 struct ocfs2_move_extents_context *context,
61 u32 cpos, u32 len, u32 p_cpos, u32 new_p_cpos,
65 struct inode *inode = context->inode;
66 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
67 struct ocfs2_extent_rec *rec, replace_rec;
68 struct ocfs2_path *path = NULL;
69 struct ocfs2_extent_list *el;
70 u64 ino = ocfs2_metadata_cache_owner(context->et.et_ci);
71 u64 old_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cpos);
73 ret = ocfs2_duplicate_clusters_by_page(handle, context->file, cpos,
74 p_cpos, new_p_cpos, len);
80 memset(&replace_rec, 0, sizeof(replace_rec));
81 replace_rec.e_cpos = cpu_to_le32(cpos);
82 replace_rec.e_leaf_clusters = cpu_to_le16(len);
83 replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(inode->i_sb,
86 path = ocfs2_new_path_from_et(&context->et);
93 ret = ocfs2_find_path(INODE_CACHE(inode), path, cpos);
99 el = path_leaf_el(path);
101 index = ocfs2_search_extent_list(el, cpos);
102 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
103 ocfs2_error(inode->i_sb,
104 "Inode %llu has an extent at cpos %u which can no "
105 "longer be found.\n",
106 (unsigned long long)ino, cpos);
111 rec = &el->l_recs[index];
113 BUG_ON(ext_flags != rec->e_flags);
115 * after moving/defraging to new location, the extent is not going
116 * to be refcounted anymore.
118 replace_rec.e_flags = ext_flags & ~OCFS2_EXT_REFCOUNTED;
120 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
121 context->et.et_root_bh,
122 OCFS2_JOURNAL_ACCESS_WRITE);
128 ret = ocfs2_split_extent(handle, &context->et, path, index,
129 &replace_rec, context->meta_ac,
136 ocfs2_journal_dirty(handle, context->et.et_root_bh);
138 context->new_phys_cpos = new_p_cpos;
141 * need I to append truncate log for old clusters?
144 if (ext_flags & OCFS2_EXT_REFCOUNTED)
145 ret = ocfs2_decrease_refcount(inode, handle,
146 ocfs2_blocks_to_clusters(osb->sb,
148 len, context->meta_ac,
149 &context->dealloc, 1);
151 ret = ocfs2_truncate_log_append(osb, handle,
160 * lock allocators, and reserving appropriate number of bits for
161 * meta blocks and data clusters.
163 * in some cases, we don't need to reserve clusters, just let data_ac
166 static int ocfs2_lock_allocators_move_extents(struct inode *inode,
167 struct ocfs2_extent_tree *et,
168 u32 clusters_to_move,
169 u32 extents_to_split,
170 struct ocfs2_alloc_context **meta_ac,
171 struct ocfs2_alloc_context **data_ac,
175 int ret, num_free_extents;
176 unsigned int max_recs_needed = 2 * extents_to_split + clusters_to_move;
177 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
179 num_free_extents = ocfs2_num_free_extents(osb, et);
180 if (num_free_extents < 0) {
181 ret = num_free_extents;
186 if (!num_free_extents ||
187 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
188 extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
190 ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, meta_ac);
197 ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac);
204 *credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el,
205 clusters_to_move + 2);
207 mlog(0, "reserve metadata_blocks: %d, data_clusters: %u, credits: %d\n",
208 extra_blocks, clusters_to_move, *credits);
212 ocfs2_free_alloc_context(*meta_ac);
221 * Using one journal handle to guarantee the data consistency in case
222 * crash happens anywhere.
224 * XXX: defrag can end up with finishing partial extent as requested,
225 * due to not enough contiguous clusters can be found in allocator.
227 static int ocfs2_defrag_extent(struct ocfs2_move_extents_context *context,
228 u32 cpos, u32 phys_cpos, u32 *len, int ext_flags)
230 int ret, credits = 0, extra_blocks = 0, partial = context->partial;
232 struct inode *inode = context->inode;
233 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
234 struct inode *tl_inode = osb->osb_tl_inode;
235 struct ocfs2_refcount_tree *ref_tree = NULL;
236 u32 new_phys_cpos, new_len;
237 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
239 if ((ext_flags & OCFS2_EXT_REFCOUNTED) && *len) {
241 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
242 OCFS2_HAS_REFCOUNT_FL));
244 BUG_ON(!context->refcount_loc);
246 ret = ocfs2_lock_refcount_tree(osb, context->refcount_loc, 1,
253 ret = ocfs2_prepare_refcount_change_for_del(inode,
254 context->refcount_loc,
265 ret = ocfs2_lock_allocators_move_extents(inode, &context->et, *len, 1,
268 extra_blocks, &credits);
275 * should be using allocation reservation strategy there?
277 * if (context->data_ac)
278 * context->data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
281 mutex_lock(&tl_inode->i_mutex);
283 if (ocfs2_truncate_log_needs_flush(osb)) {
284 ret = __ocfs2_flush_truncate_log(osb);
287 goto out_unlock_mutex;
291 handle = ocfs2_start_trans(osb, credits);
292 if (IS_ERR(handle)) {
293 ret = PTR_ERR(handle);
295 goto out_unlock_mutex;
298 ret = __ocfs2_claim_clusters(handle, context->data_ac, 1, *len,
299 &new_phys_cpos, &new_len);
306 * allowing partial extent moving is kind of 'pros and cons', it makes
307 * whole defragmentation less likely to fail, on the contrary, the bad
308 * thing is it may make the fs even more fragmented after moving, let
309 * userspace make a good decision here.
311 if (new_len != *len) {
312 mlog(0, "len_claimed: %u, len: %u\n", new_len, *len);
314 context->range->me_flags &= ~OCFS2_MOVE_EXT_FL_COMPLETE;
320 mlog(0, "cpos: %u, phys_cpos: %u, new_phys_cpos: %u\n", cpos,
321 phys_cpos, new_phys_cpos);
323 ret = __ocfs2_move_extent(handle, context, cpos, new_len, phys_cpos,
324 new_phys_cpos, ext_flags);
328 if (partial && (new_len != *len))
332 * Here we should write the new page out first if we are
333 * in write-back mode.
335 ret = ocfs2_cow_sync_writeback(inode->i_sb, context->inode, cpos, *len);
340 ocfs2_commit_trans(osb, handle);
343 mutex_unlock(&tl_inode->i_mutex);
345 if (context->data_ac) {
346 ocfs2_free_alloc_context(context->data_ac);
347 context->data_ac = NULL;
350 if (context->meta_ac) {
351 ocfs2_free_alloc_context(context->meta_ac);
352 context->meta_ac = NULL;
357 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
363 * find the victim alloc group, where #blkno fits.
365 static int ocfs2_find_victim_alloc_group(struct inode *inode,
369 struct buffer_head **ret_bh)
371 int ret, i, blocks_per_unit = 1;
375 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
376 struct buffer_head *ac_bh = NULL, *gd_bh = NULL;
377 struct ocfs2_chain_list *cl;
378 struct ocfs2_chain_rec *rec;
379 struct ocfs2_dinode *ac_dinode;
380 struct ocfs2_group_desc *bg;
382 ocfs2_sprintf_system_inode_name(namebuf, sizeof(namebuf), type, slot);
383 ret = ocfs2_lookup_ino_from_name(osb->sys_root_inode, namebuf,
384 strlen(namebuf), &blkno);
390 ret = ocfs2_read_blocks_sync(osb, blkno, 1, &ac_bh);
396 ac_dinode = (struct ocfs2_dinode *)ac_bh->b_data;
397 cl = &(ac_dinode->id2.i_chain);
398 rec = &(cl->cl_recs[0]);
400 if (type == GLOBAL_BITMAP_SYSTEM_INODE)
401 blocks_per_unit <<= (osb->s_clustersize_bits -
402 inode->i_sb->s_blocksize_bits);
404 * 'vict_blkno' was out of the valid range.
406 if ((vict_blkno < le64_to_cpu(rec->c_blkno)) ||
407 (vict_blkno >= (le32_to_cpu(ac_dinode->id1.bitmap1.i_total) *
413 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i++) {
415 rec = &(cl->cl_recs[i]);
423 blkno = le64_to_cpu(rec->c_blkno);
425 blkno = le64_to_cpu(bg->bg_next_group);
432 ret = ocfs2_read_blocks_sync(osb, blkno, 1, &gd_bh);
438 bg = (struct ocfs2_group_desc *)gd_bh->b_data;
440 if (vict_blkno < (le64_to_cpu(bg->bg_blkno) +
441 le16_to_cpu(bg->bg_bits))) {
444 *vict_bit = (vict_blkno - blkno) /
446 mlog(0, "find the victim group: #%llu, "
447 "total_bits: %u, vict_bit: %u\n",
448 blkno, le16_to_cpu(bg->bg_bits),
453 } while (le64_to_cpu(bg->bg_next_group));
461 * caller has to release the gd_bh properly.
467 * XXX: helper to validate and adjust moving goal.
469 static int ocfs2_validate_and_adjust_move_goal(struct inode *inode,
470 struct ocfs2_move_extents *range)
472 int ret, goal_bit = 0;
474 struct buffer_head *gd_bh = NULL;
475 struct ocfs2_group_desc *bg;
476 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
477 int c_to_b = 1 << (osb->s_clustersize_bits -
478 inode->i_sb->s_blocksize_bits);
481 * validate goal sits within global_bitmap, and return the victim
484 ret = ocfs2_find_victim_alloc_group(inode, range->me_goal,
485 GLOBAL_BITMAP_SYSTEM_INODE,
491 bg = (struct ocfs2_group_desc *)gd_bh->b_data;
494 * make goal become cluster aligned.
496 if (range->me_goal % c_to_b)
497 range->me_goal = range->me_goal / c_to_b * c_to_b;
500 * moving goal is not allowd to start with a group desc blok(#0 blk)
501 * let's compromise to the latter cluster.
503 if (range->me_goal == le64_to_cpu(bg->bg_blkno))
504 range->me_goal += c_to_b;
507 * movement is not gonna cross two groups.
509 if ((le16_to_cpu(bg->bg_bits) - goal_bit) * osb->s_clustersize <
515 * more exact validations/adjustments will be performed later during
516 * moving operation for each extent range.
518 mlog(0, "extents get ready to be moved to #%llu block\n",
527 static void ocfs2_probe_alloc_group(struct inode *inode, struct buffer_head *bh,
528 int *goal_bit, u32 move_len, u32 max_hop,
531 int i, used, last_free_bits = 0, base_bit = *goal_bit;
532 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
533 u32 base_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
534 le64_to_cpu(gd->bg_blkno));
536 for (i = base_bit; i < le16_to_cpu(gd->bg_bits); i++) {
538 used = ocfs2_test_bit(i, (unsigned long *)gd->bg_bitmap);
541 * we even tried searching the free chunk by jumping
542 * a 'max_hop' distance, but still failed.
544 if ((i - base_bit) > max_hop) {
556 if (last_free_bits == move_len) {
558 *phys_cpos = base_cpos + i;
563 mlog(0, "found phys_cpos: %u to fit the wanted moving.\n", *phys_cpos);
566 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
568 struct buffer_head *di_bh,
574 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
575 struct ocfs2_chain_list *cl =
576 (struct ocfs2_chain_list *) &di->id2.i_chain;
578 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
579 OCFS2_JOURNAL_ACCESS_WRITE);
585 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
586 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
587 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
588 ocfs2_journal_dirty(handle, di_bh);
594 static inline int ocfs2_block_group_set_bits(handle_t *handle,
595 struct inode *alloc_inode,
596 struct ocfs2_group_desc *bg,
597 struct buffer_head *group_bh,
598 unsigned int bit_off,
599 unsigned int num_bits)
602 void *bitmap = bg->bg_bitmap;
603 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
605 /* All callers get the descriptor via
606 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
607 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
608 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
610 mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
613 if (ocfs2_is_cluster_bitmap(alloc_inode))
614 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
616 status = ocfs2_journal_access_gd(handle,
617 INODE_CACHE(alloc_inode),
625 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
626 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
627 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
628 " count %u but claims %u are freed. num_bits %d",
629 (unsigned long long)le64_to_cpu(bg->bg_blkno),
630 le16_to_cpu(bg->bg_bits),
631 le16_to_cpu(bg->bg_free_bits_count), num_bits);
635 ocfs2_set_bit(bit_off++, bitmap);
637 ocfs2_journal_dirty(handle, group_bh);
643 static int ocfs2_move_extent(struct ocfs2_move_extents_context *context,
644 u32 cpos, u32 phys_cpos, u32 *new_phys_cpos,
645 u32 len, int ext_flags)
647 int ret, credits = 0, extra_blocks = 0, goal_bit = 0;
649 struct inode *inode = context->inode;
650 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
651 struct inode *tl_inode = osb->osb_tl_inode;
652 struct inode *gb_inode = NULL;
653 struct buffer_head *gb_bh = NULL;
654 struct buffer_head *gd_bh = NULL;
655 struct ocfs2_group_desc *gd;
656 struct ocfs2_refcount_tree *ref_tree = NULL;
657 u32 move_max_hop = ocfs2_blocks_to_clusters(inode->i_sb,
658 context->range->me_threshold);
659 u64 phys_blkno, new_phys_blkno;
661 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
663 if ((ext_flags & OCFS2_EXT_REFCOUNTED) && len) {
665 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
666 OCFS2_HAS_REFCOUNT_FL));
668 BUG_ON(!context->refcount_loc);
670 ret = ocfs2_lock_refcount_tree(osb, context->refcount_loc, 1,
677 ret = ocfs2_prepare_refcount_change_for_del(inode,
678 context->refcount_loc,
689 ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1,
691 NULL, extra_blocks, &credits);
698 * need to count 2 extra credits for global_bitmap inode and
701 credits += OCFS2_INODE_UPDATE_CREDITS + 1;
704 * ocfs2_move_extent() didn't reserve any clusters in lock_allocators()
705 * logic, while we still need to lock the global_bitmap.
707 gb_inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
710 mlog(ML_ERROR, "unable to get global_bitmap inode\n");
715 mutex_lock(&gb_inode->i_mutex);
717 ret = ocfs2_inode_lock(gb_inode, &gb_bh, 1);
720 goto out_unlock_gb_mutex;
723 mutex_lock(&tl_inode->i_mutex);
725 handle = ocfs2_start_trans(osb, credits);
726 if (IS_ERR(handle)) {
727 ret = PTR_ERR(handle);
729 goto out_unlock_tl_inode;
732 new_phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *new_phys_cpos);
733 ret = ocfs2_find_victim_alloc_group(inode, new_phys_blkno,
734 GLOBAL_BITMAP_SYSTEM_INODE,
743 * probe the victim cluster group to find a proper
744 * region to fit wanted movement, it even will perfrom
745 * a best-effort attempt by compromising to a threshold
748 ocfs2_probe_alloc_group(inode, gd_bh, &goal_bit, len, move_max_hop,
750 if (!new_phys_cpos) {
755 ret = __ocfs2_move_extent(handle, context, cpos, len, phys_cpos,
756 *new_phys_cpos, ext_flags);
762 gd = (struct ocfs2_group_desc *)gd_bh->b_data;
763 ret = ocfs2_alloc_dinode_update_counts(gb_inode, handle, gb_bh, len,
764 le16_to_cpu(gd->bg_chain));
770 ret = ocfs2_block_group_set_bits(handle, gb_inode, gd, gd_bh,
776 * Here we should write the new page out first if we are
777 * in write-back mode.
779 ret = ocfs2_cow_sync_writeback(inode->i_sb, context->inode, cpos, len);
784 ocfs2_commit_trans(osb, handle);
788 mutex_unlock(&tl_inode->i_mutex);
790 ocfs2_inode_unlock(gb_inode, 1);
792 mutex_unlock(&gb_inode->i_mutex);
797 if (context->meta_ac) {
798 ocfs2_free_alloc_context(context->meta_ac);
799 context->meta_ac = NULL;
803 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
809 * Helper to calculate the defraging length in one run according to threshold.
811 static void ocfs2_calc_extent_defrag_len(u32 *alloc_size, u32 *len_defraged,
812 u32 threshold, int *skip)
814 if ((*alloc_size + *len_defraged) < threshold) {
816 * proceed defragmentation until we meet the thresh
818 *len_defraged += *alloc_size;
819 } else if (*len_defraged == 0) {
821 * XXX: skip a large extent.
826 * split this extent to coalesce with former pieces as
827 * to reach the threshold.
829 * we're done here with one cycle of defragmentation
830 * in a size of 'thresh', resetting 'len_defraged'
831 * forces a new defragmentation.
833 *alloc_size = threshold - *len_defraged;
838 static int __ocfs2_move_extents_range(struct buffer_head *di_bh,
839 struct ocfs2_move_extents_context *context)
841 int ret = 0, flags, do_defrag, skip = 0;
842 u32 cpos, phys_cpos, move_start, len_to_move, alloc_size;
843 u32 len_defraged = 0, defrag_thresh = 0, new_phys_cpos = 0;
845 struct inode *inode = context->inode;
846 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
847 struct ocfs2_move_extents *range = context->range;
848 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
850 if ((inode->i_size == 0) || (range->me_len == 0))
853 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
856 context->refcount_loc = le64_to_cpu(di->i_refcount_loc);
858 ocfs2_init_dinode_extent_tree(&context->et, INODE_CACHE(inode), di_bh);
859 ocfs2_init_dealloc_ctxt(&context->dealloc);
867 do_defrag = context->auto_defrag;
870 * extents moving happens in unit of clusters, for the sake
871 * of simplicity, we may ignore two clusters where 'byte_start'
872 * and 'byte_start + len' were within.
874 move_start = ocfs2_clusters_for_bytes(osb->sb, range->me_start);
875 len_to_move = (range->me_start + range->me_len) >>
876 osb->s_clustersize_bits;
877 if (len_to_move >= move_start)
878 len_to_move -= move_start;
883 defrag_thresh = range->me_threshold >> osb->s_clustersize_bits;
884 if (defrag_thresh <= 1)
887 new_phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
890 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u, "
892 (unsigned long long)OCFS2_I(inode)->ip_blkno,
893 (unsigned long long)range->me_start,
894 (unsigned long long)range->me_len,
895 move_start, len_to_move, defrag_thresh);
898 while (len_to_move) {
899 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &alloc_size,
906 if (alloc_size > len_to_move)
907 alloc_size = len_to_move;
910 * XXX: how to deal with a hole:
912 * - skip the hole of course
913 * - force a new defragmentation
923 ocfs2_calc_extent_defrag_len(&alloc_size, &len_defraged,
924 defrag_thresh, &skip);
933 mlog(0, "#Defrag: cpos: %u, phys_cpos: %u, "
934 "alloc_size: %u, len_defraged: %u\n",
935 cpos, phys_cpos, alloc_size, len_defraged);
937 ret = ocfs2_defrag_extent(context, cpos, phys_cpos,
940 ret = ocfs2_move_extent(context, cpos, phys_cpos,
941 &new_phys_cpos, alloc_size,
944 new_phys_cpos += alloc_size;
952 context->clusters_moved += alloc_size;
955 len_to_move -= alloc_size;
959 range->me_flags |= OCFS2_MOVE_EXT_FL_COMPLETE;
962 range->me_moved_len = ocfs2_clusters_to_bytes(osb->sb,
963 context->clusters_moved);
964 range->me_new_offset = ocfs2_clusters_to_bytes(osb->sb,
965 context->new_phys_cpos);
967 ocfs2_schedule_truncate_log_flush(osb, 1);
968 ocfs2_run_deallocs(osb, &context->dealloc);
973 static int ocfs2_move_extents(struct ocfs2_move_extents_context *context)
977 struct inode *inode = context->inode;
978 struct ocfs2_dinode *di;
979 struct buffer_head *di_bh = NULL;
980 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
985 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
988 mutex_lock(&inode->i_mutex);
991 * This prevents concurrent writes from other nodes
993 status = ocfs2_rw_lock(inode, 1);
999 status = ocfs2_inode_lock(inode, &di_bh, 1);
1006 * rememer ip_xattr_sem also needs to be held if necessary
1008 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1010 status = __ocfs2_move_extents_range(di_bh, context);
1012 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1015 goto out_inode_unlock;
1019 * We update ctime for these changes
1021 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1022 if (IS_ERR(handle)) {
1023 status = PTR_ERR(handle);
1025 goto out_inode_unlock;
1028 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1029 OCFS2_JOURNAL_ACCESS_WRITE);
1035 di = (struct ocfs2_dinode *)di_bh->b_data;
1036 inode->i_ctime = CURRENT_TIME;
1037 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1038 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1040 ocfs2_journal_dirty(handle, di_bh);
1043 ocfs2_commit_trans(osb, handle);
1047 ocfs2_inode_unlock(inode, 1);
1049 ocfs2_rw_unlock(inode, 1);
1051 mutex_unlock(&inode->i_mutex);
1056 int ocfs2_ioctl_move_extents(struct file *filp, void __user *argp)
1060 struct inode *inode = filp->f_path.dentry->d_inode;
1061 struct ocfs2_move_extents range;
1062 struct ocfs2_move_extents_context *context = NULL;
1064 status = mnt_want_write(filp->f_path.mnt);
1068 if ((!S_ISREG(inode->i_mode)) || !(filp->f_mode & FMODE_WRITE))
1071 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1076 context = kzalloc(sizeof(struct ocfs2_move_extents_context), GFP_NOFS);
1083 context->inode = inode;
1084 context->file = filp;
1087 if (copy_from_user(&range, (struct ocfs2_move_extents *)argp,
1097 if (range.me_start > i_size_read(inode))
1100 if (range.me_start + range.me_len > i_size_read(inode))
1101 range.me_len = i_size_read(inode) - range.me_start;
1103 context->range = ⦥
1105 if (range.me_flags & OCFS2_MOVE_EXT_FL_AUTO_DEFRAG) {
1106 context->auto_defrag = 1;
1108 * ok, the default theshold for the defragmentation
1109 * is 1M, since our maximum clustersize was 1M also.
1112 if (!range.me_threshold)
1113 range.me_threshold = 1024 * 1024;
1115 if (range.me_threshold > i_size_read(inode))
1116 range.me_threshold = i_size_read(inode);
1118 if (range.me_flags & OCFS2_MOVE_EXT_FL_PART_DEFRAG)
1119 context->partial = 1;
1122 * first best-effort attempt to validate and adjust the goal
1123 * (physical address in block), while it can't guarantee later
1124 * operation can succeed all the time since global_bitmap may
1125 * change a bit over time.
1128 status = ocfs2_validate_and_adjust_move_goal(inode, &range);
1133 status = ocfs2_move_extents(context);
1138 * movement/defragmentation may end up being partially completed,
1139 * that's the reason why we need to return userspace the finished
1140 * length and new_offset even if failure happens somewhere.
1143 if (copy_to_user((struct ocfs2_move_extents *)argp, &range,
1150 mnt_drop_write(filp->f_path.mnt);