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 {
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, context->file, 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);
101 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
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,
159 * lock allocators, and reserving appropriate number of bits for
160 * meta blocks and data clusters.
162 * in some cases, we don't need to reserve clusters, just let data_ac
165 static int ocfs2_lock_allocators_move_extents(struct inode *inode,
166 struct ocfs2_extent_tree *et,
167 u32 clusters_to_move,
168 u32 extents_to_split,
169 struct ocfs2_alloc_context **meta_ac,
170 struct ocfs2_alloc_context **data_ac,
174 int ret, num_free_extents;
175 unsigned int max_recs_needed = 2 * extents_to_split + clusters_to_move;
176 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
178 num_free_extents = ocfs2_num_free_extents(osb, et);
179 if (num_free_extents < 0) {
180 ret = num_free_extents;
185 if (!num_free_extents ||
186 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
187 extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
189 ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, meta_ac);
196 ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac);
203 *credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el,
204 clusters_to_move + 2);
206 mlog(0, "reserve metadata_blocks: %d, data_clusters: %u, credits: %d\n",
207 extra_blocks, clusters_to_move, *credits);
211 ocfs2_free_alloc_context(*meta_ac);
220 * Using one journal handle to guarantee the data consistency in case
221 * crash happens anywhere.
223 static int ocfs2_defrag_extent(struct ocfs2_move_extents_context *context,
224 u32 cpos, u32 phys_cpos, u32 len, int ext_flags)
226 int ret, credits = 0, extra_blocks = 0;
228 struct inode *inode = context->inode;
229 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
230 struct inode *tl_inode = osb->osb_tl_inode;
231 struct ocfs2_refcount_tree *ref_tree = NULL;
232 u32 new_phys_cpos, new_len;
233 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
235 if ((ext_flags & OCFS2_EXT_REFCOUNTED) && len) {
237 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
238 OCFS2_HAS_REFCOUNT_FL));
240 BUG_ON(!context->refcount_loc);
242 ret = ocfs2_lock_refcount_tree(osb, context->refcount_loc, 1,
249 ret = ocfs2_prepare_refcount_change_for_del(inode,
250 context->refcount_loc,
261 ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1,
264 extra_blocks, &credits);
271 * should be using allocation reservation strategy there?
273 * if (context->data_ac)
274 * context->data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
277 mutex_lock(&tl_inode->i_mutex);
279 if (ocfs2_truncate_log_needs_flush(osb)) {
280 ret = __ocfs2_flush_truncate_log(osb);
283 goto out_unlock_mutex;
287 handle = ocfs2_start_trans(osb, credits);
288 if (IS_ERR(handle)) {
289 ret = PTR_ERR(handle);
291 goto out_unlock_mutex;
294 ret = __ocfs2_claim_clusters(handle, context->data_ac, 1, len,
295 &new_phys_cpos, &new_len);
302 * we're not quite patient here to make multiple attempts for claiming
303 * enough clusters, failure to claim clusters per-requested is not a
304 * disaster though, it can only mean partial range of defragmentation
305 * or extent movements gets gone, users anyway is able to have another
306 * try as they wish anytime, since they're going to be returned a
307 * '-ENOSPC' and completed length of this movement.
309 if (new_len != len) {
310 mlog(0, "len_claimed: %u, len: %u\n", new_len, len);
311 context->range->me_flags &= ~OCFS2_MOVE_EXT_FL_COMPLETE;
316 mlog(0, "cpos: %u, phys_cpos: %u, new_phys_cpos: %u\n", cpos,
317 phys_cpos, new_phys_cpos);
319 ret = __ocfs2_move_extent(handle, context, cpos, len, phys_cpos,
320 new_phys_cpos, ext_flags);
325 * Here we should write the new page out first if we are
326 * in write-back mode.
328 ret = ocfs2_cow_sync_writeback(inode->i_sb, context->inode, cpos, len);
333 ocfs2_commit_trans(osb, handle);
336 mutex_unlock(&tl_inode->i_mutex);
338 if (context->data_ac) {
339 ocfs2_free_alloc_context(context->data_ac);
340 context->data_ac = NULL;
343 if (context->meta_ac) {
344 ocfs2_free_alloc_context(context->meta_ac);
345 context->meta_ac = NULL;
350 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
356 * find the victim alloc group, where #blkno fits.
358 static int ocfs2_find_victim_alloc_group(struct inode *inode,
362 struct buffer_head **ret_bh)
364 int ret, i, blocks_per_unit = 1;
368 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
369 struct buffer_head *ac_bh = NULL, *gd_bh = NULL;
370 struct ocfs2_chain_list *cl;
371 struct ocfs2_chain_rec *rec;
372 struct ocfs2_dinode *ac_dinode;
373 struct ocfs2_group_desc *bg;
375 ocfs2_sprintf_system_inode_name(namebuf, sizeof(namebuf), type, slot);
376 ret = ocfs2_lookup_ino_from_name(osb->sys_root_inode, namebuf,
377 strlen(namebuf), &blkno);
383 ret = ocfs2_read_blocks_sync(osb, blkno, 1, &ac_bh);
389 ac_dinode = (struct ocfs2_dinode *)ac_bh->b_data;
390 cl = &(ac_dinode->id2.i_chain);
391 rec = &(cl->cl_recs[0]);
393 if (type == GLOBAL_BITMAP_SYSTEM_INODE)
394 blocks_per_unit <<= (osb->s_clustersize_bits -
395 inode->i_sb->s_blocksize_bits);
397 * 'vict_blkno' was out of the valid range.
399 if ((vict_blkno < le64_to_cpu(rec->c_blkno)) ||
400 (vict_blkno >= (le32_to_cpu(ac_dinode->id1.bitmap1.i_total) *
406 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i++) {
408 rec = &(cl->cl_recs[i]);
416 blkno = le64_to_cpu(rec->c_blkno);
418 blkno = le64_to_cpu(bg->bg_next_group);
425 ret = ocfs2_read_blocks_sync(osb, blkno, 1, &gd_bh);
431 bg = (struct ocfs2_group_desc *)gd_bh->b_data;
433 if (vict_blkno < (le64_to_cpu(bg->bg_blkno) +
434 le16_to_cpu(bg->bg_bits))) {
437 *vict_bit = (vict_blkno - blkno) /
439 mlog(0, "find the victim group: #%llu, "
440 "total_bits: %u, vict_bit: %u\n",
441 blkno, le16_to_cpu(bg->bg_bits),
446 } while (le64_to_cpu(bg->bg_next_group));
454 * caller has to release the gd_bh properly.
460 * XXX: helper to validate and adjust moving goal.
462 static int ocfs2_validate_and_adjust_move_goal(struct inode *inode,
463 struct ocfs2_move_extents *range)
465 int ret, goal_bit = 0;
467 struct buffer_head *gd_bh = NULL;
468 struct ocfs2_group_desc *bg;
469 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
470 int c_to_b = 1 << (osb->s_clustersize_bits -
471 inode->i_sb->s_blocksize_bits);
474 * validate goal sits within global_bitmap, and return the victim
477 ret = ocfs2_find_victim_alloc_group(inode, range->me_goal,
478 GLOBAL_BITMAP_SYSTEM_INODE,
484 bg = (struct ocfs2_group_desc *)gd_bh->b_data;
487 * make goal become cluster aligned.
489 if (range->me_goal % c_to_b)
490 range->me_goal = range->me_goal / c_to_b * c_to_b;
493 * moving goal is not allowd to start with a group desc blok(#0 blk)
494 * let's compromise to the latter cluster.
496 if (range->me_goal == le64_to_cpu(bg->bg_blkno))
497 range->me_goal += c_to_b;
500 * movement is not gonna cross two groups.
502 if ((le16_to_cpu(bg->bg_bits) - goal_bit) * osb->s_clustersize <
508 * more exact validations/adjustments will be performed later during
509 * moving operation for each extent range.
511 mlog(0, "extents get ready to be moved to #%llu block\n",
520 static void ocfs2_probe_alloc_group(struct inode *inode, struct buffer_head *bh,
521 int *goal_bit, u32 move_len, u32 max_hop,
524 int i, used, last_free_bits = 0, base_bit = *goal_bit;
525 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
526 u32 base_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
527 le64_to_cpu(gd->bg_blkno));
529 for (i = base_bit; i < le16_to_cpu(gd->bg_bits); i++) {
531 used = ocfs2_test_bit(i, (unsigned long *)gd->bg_bitmap);
534 * we even tried searching the free chunk by jumping
535 * a 'max_hop' distance, but still failed.
537 if ((i - base_bit) > max_hop) {
549 if (last_free_bits == move_len) {
551 *phys_cpos = base_cpos + i;
556 mlog(0, "found phys_cpos: %u to fit the wanted moving.\n", *phys_cpos);
559 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
561 struct buffer_head *di_bh,
567 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
568 struct ocfs2_chain_list *cl =
569 (struct ocfs2_chain_list *) &di->id2.i_chain;
571 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
572 OCFS2_JOURNAL_ACCESS_WRITE);
578 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
579 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
580 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
581 ocfs2_journal_dirty(handle, di_bh);
587 static inline int ocfs2_block_group_set_bits(handle_t *handle,
588 struct inode *alloc_inode,
589 struct ocfs2_group_desc *bg,
590 struct buffer_head *group_bh,
591 unsigned int bit_off,
592 unsigned int num_bits)
595 void *bitmap = bg->bg_bitmap;
596 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
598 /* All callers get the descriptor via
599 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
600 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
601 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
603 mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
606 if (ocfs2_is_cluster_bitmap(alloc_inode))
607 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
609 status = ocfs2_journal_access_gd(handle,
610 INODE_CACHE(alloc_inode),
618 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
619 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
620 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
621 " count %u but claims %u are freed. num_bits %d",
622 (unsigned long long)le64_to_cpu(bg->bg_blkno),
623 le16_to_cpu(bg->bg_bits),
624 le16_to_cpu(bg->bg_free_bits_count), num_bits);
628 ocfs2_set_bit(bit_off++, bitmap);
630 ocfs2_journal_dirty(handle, group_bh);