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"
39 #include "refcounttree.h"
40 #include "move_extents.h"
42 struct ocfs2_move_extents_context {
51 struct ocfs2_move_extents *range;
52 struct ocfs2_extent_tree et;
53 struct ocfs2_alloc_context *meta_ac;
54 struct ocfs2_alloc_context *data_ac;
55 struct ocfs2_cached_dealloc_ctxt dealloc;
58 static int __ocfs2_move_extent(handle_t *handle,
59 struct ocfs2_move_extents_context *context,
60 u32 cpos, u32 len, u32 p_cpos, u32 new_p_cpos,
64 struct inode *inode = context->inode;
65 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
66 struct ocfs2_extent_rec *rec, replace_rec;
67 struct ocfs2_path *path = NULL;
68 struct ocfs2_extent_list *el;
69 u64 ino = ocfs2_metadata_cache_owner(context->et.et_ci);
70 u64 old_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cpos);
72 ret = ocfs2_duplicate_clusters_by_page(handle, inode, cpos,
73 p_cpos, new_p_cpos, len);
79 memset(&replace_rec, 0, sizeof(replace_rec));
80 replace_rec.e_cpos = cpu_to_le32(cpos);
81 replace_rec.e_leaf_clusters = cpu_to_le16(len);
82 replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(inode->i_sb,
85 path = ocfs2_new_path_from_et(&context->et);
92 ret = ocfs2_find_path(INODE_CACHE(inode), path, cpos);
98 el = path_leaf_el(path);
100 index = ocfs2_search_extent_list(el, cpos);
102 ocfs2_error(inode->i_sb,
103 "Inode %llu has an extent at cpos %u which can no "
104 "longer be found.\n",
105 (unsigned long long)ino, cpos);
110 rec = &el->l_recs[index];
112 BUG_ON(ext_flags != rec->e_flags);
114 * after moving/defraging to new location, the extent is not going
115 * to be refcounted anymore.
117 replace_rec.e_flags = ext_flags & ~OCFS2_EXT_REFCOUNTED;
119 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
120 context->et.et_root_bh,
121 OCFS2_JOURNAL_ACCESS_WRITE);
127 ret = ocfs2_split_extent(handle, &context->et, path, index,
128 &replace_rec, context->meta_ac,
135 ocfs2_journal_dirty(handle, context->et.et_root_bh);
137 context->new_phys_cpos = new_p_cpos;
140 * need I to append truncate log for old clusters?
143 if (ext_flags & OCFS2_EXT_REFCOUNTED)
144 ret = ocfs2_decrease_refcount(inode, handle,
145 ocfs2_blocks_to_clusters(osb->sb,
147 len, context->meta_ac,
148 &context->dealloc, 1);
150 ret = ocfs2_truncate_log_append(osb, handle,
154 ocfs2_update_inode_fsync_trans(handle, inode, 0);
156 ocfs2_free_path(path);
161 * lock allocators, and reserving appropriate number of bits for
162 * meta blocks and data clusters.
164 * in some cases, we don't need to reserve clusters, just let data_ac
167 static int ocfs2_lock_allocators_move_extents(struct inode *inode,
168 struct ocfs2_extent_tree *et,
169 u32 clusters_to_move,
170 u32 extents_to_split,
171 struct ocfs2_alloc_context **meta_ac,
172 struct ocfs2_alloc_context **data_ac,
176 int ret, num_free_extents;
177 unsigned int max_recs_needed = 2 * extents_to_split + clusters_to_move;
178 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
180 num_free_extents = ocfs2_num_free_extents(osb, et);
181 if (num_free_extents < 0) {
182 ret = num_free_extents;
187 if (!num_free_extents ||
188 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
189 extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
191 ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, meta_ac);
198 ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac);
205 *credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el);
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, bits_per_unit = 0;
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 bits_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 * make goal become cluster aligned.
483 range->me_goal = ocfs2_block_to_cluster_start(inode->i_sb,
486 * validate goal sits within global_bitmap, and return the victim
489 ret = ocfs2_find_victim_alloc_group(inode, range->me_goal,
490 GLOBAL_BITMAP_SYSTEM_INODE,
496 bg = (struct ocfs2_group_desc *)gd_bh->b_data;
499 * moving goal is not allowd to start with a group desc blok(#0 blk)
500 * let's compromise to the latter cluster.
502 if (range->me_goal == le64_to_cpu(bg->bg_blkno))
503 range->me_goal += c_to_b;
506 * movement is not gonna cross two groups.
508 if ((le16_to_cpu(bg->bg_bits) - goal_bit) * osb->s_clustersize <
514 * more exact validations/adjustments will be performed later during
515 * moving operation for each extent range.
517 mlog(0, "extents get ready to be moved to #%llu block\n",
526 static void ocfs2_probe_alloc_group(struct inode *inode, struct buffer_head *bh,
527 int *goal_bit, u32 move_len, u32 max_hop,
530 int i, used, last_free_bits = 0, base_bit = *goal_bit;
531 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
532 u32 base_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
533 le64_to_cpu(gd->bg_blkno));
535 for (i = base_bit; i < le16_to_cpu(gd->bg_bits); i++) {
537 used = ocfs2_test_bit(i, (unsigned long *)gd->bg_bitmap);
540 * we even tried searching the free chunk by jumping
541 * a 'max_hop' distance, but still failed.
543 if ((i - base_bit) > max_hop) {
555 if (last_free_bits == move_len) {
557 *phys_cpos = base_cpos + i;
562 mlog(0, "found phys_cpos: %u to fit the wanted moving.\n", *phys_cpos);
565 static int ocfs2_move_extent(struct ocfs2_move_extents_context *context,
566 u32 cpos, u32 phys_cpos, u32 *new_phys_cpos,
567 u32 len, int ext_flags)
569 int ret, credits = 0, extra_blocks = 0, goal_bit = 0;
571 struct inode *inode = context->inode;
572 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
573 struct inode *tl_inode = osb->osb_tl_inode;
574 struct inode *gb_inode = NULL;
575 struct buffer_head *gb_bh = NULL;
576 struct buffer_head *gd_bh = NULL;
577 struct ocfs2_group_desc *gd;
578 struct ocfs2_refcount_tree *ref_tree = NULL;
579 u32 move_max_hop = ocfs2_blocks_to_clusters(inode->i_sb,
580 context->range->me_threshold);
581 u64 phys_blkno, new_phys_blkno;
583 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
585 if ((ext_flags & OCFS2_EXT_REFCOUNTED) && len) {
587 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
588 OCFS2_HAS_REFCOUNT_FL));
590 BUG_ON(!context->refcount_loc);
592 ret = ocfs2_lock_refcount_tree(osb, context->refcount_loc, 1,
599 ret = ocfs2_prepare_refcount_change_for_del(inode,
600 context->refcount_loc,
611 ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1,
613 NULL, extra_blocks, &credits);
620 * need to count 2 extra credits for global_bitmap inode and
623 credits += OCFS2_INODE_UPDATE_CREDITS + 1;
626 * ocfs2_move_extent() didn't reserve any clusters in lock_allocators()
627 * logic, while we still need to lock the global_bitmap.
629 gb_inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
632 mlog(ML_ERROR, "unable to get global_bitmap inode\n");
637 mutex_lock(&gb_inode->i_mutex);
639 ret = ocfs2_inode_lock(gb_inode, &gb_bh, 1);
642 goto out_unlock_gb_mutex;
645 mutex_lock(&tl_inode->i_mutex);
647 handle = ocfs2_start_trans(osb, credits);
648 if (IS_ERR(handle)) {
649 ret = PTR_ERR(handle);
651 goto out_unlock_tl_inode;
654 new_phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *new_phys_cpos);
655 ret = ocfs2_find_victim_alloc_group(inode, new_phys_blkno,
656 GLOBAL_BITMAP_SYSTEM_INODE,
665 * probe the victim cluster group to find a proper
666 * region to fit wanted movement, it even will perfrom
667 * a best-effort attempt by compromising to a threshold
670 ocfs2_probe_alloc_group(inode, gd_bh, &goal_bit, len, move_max_hop,
672 if (!*new_phys_cpos) {
677 ret = __ocfs2_move_extent(handle, context, cpos, len, phys_cpos,
678 *new_phys_cpos, ext_flags);
684 gd = (struct ocfs2_group_desc *)gd_bh->b_data;
685 ret = ocfs2_alloc_dinode_update_counts(gb_inode, handle, gb_bh, len,
686 le16_to_cpu(gd->bg_chain));
692 ret = ocfs2_block_group_set_bits(handle, gb_inode, gd, gd_bh,
695 ocfs2_rollback_alloc_dinode_counts(gb_inode, gb_bh, len,
696 le16_to_cpu(gd->bg_chain));
701 * Here we should write the new page out first if we are
702 * in write-back mode.
704 ret = ocfs2_cow_sync_writeback(inode->i_sb, context->inode, cpos, len);
709 ocfs2_commit_trans(osb, handle);
713 mutex_unlock(&tl_inode->i_mutex);
715 ocfs2_inode_unlock(gb_inode, 1);
717 mutex_unlock(&gb_inode->i_mutex);
722 if (context->meta_ac) {
723 ocfs2_free_alloc_context(context->meta_ac);
724 context->meta_ac = NULL;
728 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
734 * Helper to calculate the defraging length in one run according to threshold.
736 static void ocfs2_calc_extent_defrag_len(u32 *alloc_size, u32 *len_defraged,
737 u32 threshold, int *skip)
739 if ((*alloc_size + *len_defraged) < threshold) {
741 * proceed defragmentation until we meet the thresh
743 *len_defraged += *alloc_size;
744 } else if (*len_defraged == 0) {
746 * XXX: skip a large extent.
751 * split this extent to coalesce with former pieces as
752 * to reach the threshold.
754 * we're done here with one cycle of defragmentation
755 * in a size of 'thresh', resetting 'len_defraged'
756 * forces a new defragmentation.
758 *alloc_size = threshold - *len_defraged;
763 static int __ocfs2_move_extents_range(struct buffer_head *di_bh,
764 struct ocfs2_move_extents_context *context)
766 int ret = 0, flags, do_defrag, skip = 0;
767 u32 cpos, phys_cpos, move_start, len_to_move, alloc_size;
768 u32 len_defraged = 0, defrag_thresh = 0, new_phys_cpos = 0;
770 struct inode *inode = context->inode;
771 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
772 struct ocfs2_move_extents *range = context->range;
773 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
775 if ((i_size_read(inode) == 0) || (range->me_len == 0))
778 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
781 context->refcount_loc = le64_to_cpu(di->i_refcount_loc);
783 ocfs2_init_dinode_extent_tree(&context->et, INODE_CACHE(inode), di_bh);
784 ocfs2_init_dealloc_ctxt(&context->dealloc);
792 do_defrag = context->auto_defrag;
795 * extents moving happens in unit of clusters, for the sake
796 * of simplicity, we may ignore two clusters where 'byte_start'
797 * and 'byte_start + len' were within.
799 move_start = ocfs2_clusters_for_bytes(osb->sb, range->me_start);
800 len_to_move = (range->me_start + range->me_len) >>
801 osb->s_clustersize_bits;
802 if (len_to_move >= move_start)
803 len_to_move -= move_start;
808 defrag_thresh = range->me_threshold >> osb->s_clustersize_bits;
809 if (defrag_thresh <= 1)
812 new_phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
815 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u, "
817 (unsigned long long)OCFS2_I(inode)->ip_blkno,
818 (unsigned long long)range->me_start,
819 (unsigned long long)range->me_len,
820 move_start, len_to_move, defrag_thresh);
823 while (len_to_move) {
824 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &alloc_size,
831 if (alloc_size > len_to_move)
832 alloc_size = len_to_move;
835 * XXX: how to deal with a hole:
837 * - skip the hole of course
838 * - force a new defragmentation
848 ocfs2_calc_extent_defrag_len(&alloc_size, &len_defraged,
849 defrag_thresh, &skip);
858 mlog(0, "#Defrag: cpos: %u, phys_cpos: %u, "
859 "alloc_size: %u, len_defraged: %u\n",
860 cpos, phys_cpos, alloc_size, len_defraged);
862 ret = ocfs2_defrag_extent(context, cpos, phys_cpos,
865 ret = ocfs2_move_extent(context, cpos, phys_cpos,
866 &new_phys_cpos, alloc_size,
869 new_phys_cpos += alloc_size;
877 context->clusters_moved += alloc_size;
880 len_to_move -= alloc_size;
884 range->me_flags |= OCFS2_MOVE_EXT_FL_COMPLETE;
887 range->me_moved_len = ocfs2_clusters_to_bytes(osb->sb,
888 context->clusters_moved);
889 range->me_new_offset = ocfs2_clusters_to_bytes(osb->sb,
890 context->new_phys_cpos);
892 ocfs2_schedule_truncate_log_flush(osb, 1);
893 ocfs2_run_deallocs(osb, &context->dealloc);
898 static int ocfs2_move_extents(struct ocfs2_move_extents_context *context)
902 struct inode *inode = context->inode;
903 struct ocfs2_dinode *di;
904 struct buffer_head *di_bh = NULL;
905 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
910 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
913 mutex_lock(&inode->i_mutex);
916 * This prevents concurrent writes from other nodes
918 status = ocfs2_rw_lock(inode, 1);
924 status = ocfs2_inode_lock(inode, &di_bh, 1);
931 * rememer ip_xattr_sem also needs to be held if necessary
933 down_write(&OCFS2_I(inode)->ip_alloc_sem);
935 status = __ocfs2_move_extents_range(di_bh, context);
937 up_write(&OCFS2_I(inode)->ip_alloc_sem);
940 goto out_inode_unlock;
944 * We update ctime for these changes
946 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
947 if (IS_ERR(handle)) {
948 status = PTR_ERR(handle);
950 goto out_inode_unlock;
953 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
954 OCFS2_JOURNAL_ACCESS_WRITE);
960 di = (struct ocfs2_dinode *)di_bh->b_data;
961 inode->i_ctime = CURRENT_TIME;
962 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
963 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
964 ocfs2_update_inode_fsync_trans(handle, inode, 0);
966 ocfs2_journal_dirty(handle, di_bh);
969 ocfs2_commit_trans(osb, handle);
973 ocfs2_inode_unlock(inode, 1);
975 ocfs2_rw_unlock(inode, 1);
977 mutex_unlock(&inode->i_mutex);
982 int ocfs2_ioctl_move_extents(struct file *filp, void __user *argp)
986 struct inode *inode = file_inode(filp);
987 struct ocfs2_move_extents range;
988 struct ocfs2_move_extents_context *context;
993 status = mnt_want_write_file(filp);
997 if ((!S_ISREG(inode->i_mode)) || !(filp->f_mode & FMODE_WRITE)) {
1002 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1007 context = kzalloc(sizeof(struct ocfs2_move_extents_context), GFP_NOFS);
1014 context->inode = inode;
1015 context->file = filp;
1017 if (copy_from_user(&range, argp, sizeof(range))) {
1022 if (range.me_start > i_size_read(inode)) {
1027 if (range.me_start + range.me_len > i_size_read(inode))
1028 range.me_len = i_size_read(inode) - range.me_start;
1030 context->range = ⦥
1032 if (range.me_flags & OCFS2_MOVE_EXT_FL_AUTO_DEFRAG) {
1033 context->auto_defrag = 1;
1035 * ok, the default theshold for the defragmentation
1036 * is 1M, since our maximum clustersize was 1M also.
1039 if (!range.me_threshold)
1040 range.me_threshold = 1024 * 1024;
1042 if (range.me_threshold > i_size_read(inode))
1043 range.me_threshold = i_size_read(inode);
1045 if (range.me_flags & OCFS2_MOVE_EXT_FL_PART_DEFRAG)
1046 context->partial = 1;
1049 * first best-effort attempt to validate and adjust the goal
1050 * (physical address in block), while it can't guarantee later
1051 * operation can succeed all the time since global_bitmap may
1052 * change a bit over time.
1055 status = ocfs2_validate_and_adjust_move_goal(inode, &range);
1060 status = ocfs2_move_extents(context);
1065 * movement/defragmentation may end up being partially completed,
1066 * that's the reason why we need to return userspace the finished
1067 * length and new_offset even if failure happens somewhere.
1069 if (copy_to_user(argp, &range, sizeof(range)))
1075 mnt_drop_write_file(filp);