2 * Copyright (c) 2008 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/vfs/hammer/hammer_reblock.c,v 1.17 2008/06/09 04:19:10 dillon Exp $
37 * HAMMER reblocker - This code frees up fragmented physical space
39 * HAMMER only keeps track of free space on a big-block basis. A big-block
40 * containing holes can only be freed by migrating the remaining data in
41 * that big-block into a new big-block, then freeing the big-block.
43 * This function is called from an ioctl or via the hammer support thread.
48 static int hammer_reblock_helper(struct hammer_ioc_reblock *reblock,
49 hammer_cursor_t cursor,
50 hammer_btree_elm_t elm);
51 static int hammer_reblock_data(struct hammer_ioc_reblock *reblock,
52 hammer_cursor_t cursor, hammer_btree_elm_t elm);
53 static int hammer_reblock_leaf_node(struct hammer_ioc_reblock *reblock,
54 hammer_cursor_t cursor, hammer_btree_elm_t elm);
55 static int hammer_reblock_int_node(struct hammer_ioc_reblock *reblock,
56 hammer_cursor_t cursor, hammer_btree_elm_t elm);
59 hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
60 struct hammer_ioc_reblock *reblock)
62 struct hammer_cursor cursor;
63 hammer_btree_elm_t elm;
66 if (reblock->beg_obj_id >= reblock->end_obj_id)
68 if (reblock->free_level < 0)
71 reblock->cur_obj_id = reblock->beg_obj_id;
72 reblock->cur_localization = reblock->beg_localization;
75 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
77 hammer_done_cursor(&cursor);
80 cursor.key_beg.localization = reblock->cur_localization;
81 cursor.key_beg.obj_id = reblock->cur_obj_id;
82 cursor.key_beg.key = HAMMER_MIN_KEY;
83 cursor.key_beg.create_tid = 1;
84 cursor.key_beg.delete_tid = 0;
85 cursor.key_beg.rec_type = HAMMER_MIN_RECTYPE;
86 cursor.key_beg.obj_type = 0;
88 cursor.key_end.localization = reblock->end_localization;
89 cursor.key_end.obj_id = reblock->end_obj_id;
90 cursor.key_end.key = HAMMER_MAX_KEY;
91 cursor.key_end.create_tid = HAMMER_MAX_TID - 1;
92 cursor.key_end.delete_tid = 0;
93 cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
94 cursor.key_end.obj_type = 0;
96 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
97 cursor.flags |= HAMMER_CURSOR_BACKEND;
100 * This flag allows the btree scan code to return internal nodes,
101 * so we can reblock them in addition to the leafs. Only specify it
102 * if we intend to reblock B-Tree nodes.
104 if (reblock->head.flags & HAMMER_IOC_DO_BTREE)
105 cursor.flags |= HAMMER_CURSOR_REBLOCKING;
107 error = hammer_btree_first(&cursor);
110 * Internal or Leaf node
112 elm = &cursor.node->ondisk->elms[cursor.index];
113 reblock->cur_obj_id = elm->base.obj_id;
114 reblock->cur_localization = elm->base.localization;
117 * Yield to more important tasks
119 if ((error = hammer_signal_check(trans->hmp)) != 0)
121 if (trans->hmp->sync_lock.wanted) {
122 tsleep(trans, 0, "hmrslo", hz / 10);
124 if (trans->hmp->locked_dirty_count +
125 trans->hmp->io_running_count > hammer_limit_dirtybufs) {
126 hammer_flusher_async(trans->hmp);
127 tsleep(trans, 0, "hmrslo", hz / 10);
131 * Acquiring the sync_lock prevents the operation from
132 * crossing a synchronization boundary.
134 * NOTE: cursor.node may have changed on return.
136 hammer_sync_lock_sh(trans);
137 error = hammer_reblock_helper(reblock, &cursor, elm);
138 hammer_sync_unlock(trans);
140 cursor.flags |= HAMMER_CURSOR_ATEDISK;
141 error = hammer_btree_iterate(&cursor);
146 hammer_done_cursor(&cursor);
147 if (error == EDEADLK)
149 if (error == EINTR) {
150 reblock->head.flags |= HAMMER_IOC_HEAD_INTR;
157 * Reblock the B-Tree (leaf) node, record, and/or data if necessary.
159 * XXX We have no visibility into internal B-Tree nodes at the moment,
163 hammer_reblock_helper(struct hammer_ioc_reblock *reblock,
164 hammer_cursor_t cursor, hammer_btree_elm_t elm)
166 hammer_off_t tmp_offset;
175 * Reblock data. Note that data embedded in a record is reblocked
176 * by the record reblock code. Data processing only occurs at leaf
177 * nodes and for RECORD element types.
179 if (cursor->node->ondisk->type != HAMMER_BTREE_TYPE_LEAF)
181 if (elm->leaf.base.btype != HAMMER_BTREE_TYPE_RECORD)
183 tmp_offset = elm->leaf.data_offset;
184 zone = HAMMER_ZONE_DECODE(tmp_offset); /* can be 0 */
185 if ((zone == HAMMER_ZONE_SMALL_DATA_INDEX ||
186 zone == HAMMER_ZONE_LARGE_DATA_INDEX) &&
187 error == 0 && (reblock->head.flags & (HAMMER_IOC_DO_DATA | HAMMER_IOC_DO_INODES))) {
188 ++reblock->data_count;
189 reblock->data_byte_count += elm->leaf.data_len;
190 bytes = hammer_blockmap_getfree(cursor->trans->hmp, tmp_offset,
192 if (hammer_debug_general & 0x4000)
193 kprintf("D %6d/%d\n", bytes, reblock->free_level);
194 if (error == 0 && cur == 0 && bytes >= reblock->free_level) {
195 error = hammer_cursor_upgrade(cursor);
197 error = hammer_reblock_data(reblock,
201 ++reblock->data_moves;
202 reblock->data_byte_moves += elm->leaf.data_len;
209 * Reblock a B-Tree internal or leaf node.
211 tmp_offset = cursor->node->node_offset;
212 zone = HAMMER_ZONE_DECODE(tmp_offset);
213 if (zone == HAMMER_ZONE_BTREE_INDEX && cursor->index == 0 &&
214 error == 0 && (reblock->head.flags & HAMMER_IOC_DO_BTREE)) {
215 ++reblock->btree_count;
216 bytes = hammer_blockmap_getfree(cursor->trans->hmp, tmp_offset,
218 if (hammer_debug_general & 0x4000)
219 kprintf("B %6d/%d\n", bytes, reblock->free_level);
220 if (error == 0 && cur == 0 && bytes >= reblock->free_level) {
221 error = hammer_cursor_upgrade(cursor);
224 elm = &cursor->parent->ondisk->elms[cursor->parent_index];
227 switch(cursor->node->ondisk->type) {
228 case HAMMER_BTREE_TYPE_LEAF:
229 error = hammer_reblock_leaf_node(
230 reblock, cursor, elm);
232 case HAMMER_BTREE_TYPE_INTERNAL:
233 error = hammer_reblock_int_node(
234 reblock, cursor, elm);
237 panic("Illegal B-Tree node type");
241 ++reblock->btree_moves;
246 hammer_cursor_downgrade(cursor);
251 * Reblock a record's data. Both the B-Tree element and record pointers
252 * to the data must be adjusted.
255 hammer_reblock_data(struct hammer_ioc_reblock *reblock,
256 hammer_cursor_t cursor, hammer_btree_elm_t elm)
258 struct hammer_buffer *data_buffer = NULL;
259 hammer_off_t ndata_offset;
263 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA |
264 HAMMER_CURSOR_GET_LEAF);
267 ndata = hammer_alloc_data(cursor->trans, elm->leaf.data_len,
268 &ndata_offset, &data_buffer, &error);
275 hammer_modify_buffer(cursor->trans, data_buffer, NULL, 0);
276 bcopy(cursor->data, ndata, elm->leaf.data_len);
277 hammer_modify_buffer_done(data_buffer);
279 hammer_blockmap_free(cursor->trans,
280 elm->leaf.data_offset, elm->leaf.data_len);
282 hammer_modify_node(cursor->trans, cursor->node,
283 &elm->leaf.data_offset, sizeof(hammer_off_t));
284 elm->leaf.data_offset = ndata_offset;
285 hammer_modify_node_done(cursor->node);
289 hammer_rel_buffer(data_buffer, 0);
294 * Reblock a B-Tree leaf node. The parent must be adjusted to point to
295 * the new copy of the leaf node.
297 * elm is a pointer to the parent element pointing at cursor.node.
300 hammer_reblock_leaf_node(struct hammer_ioc_reblock *reblock,
301 hammer_cursor_t cursor, hammer_btree_elm_t elm)
307 onode = cursor->node;
308 nnode = hammer_alloc_btree(cursor->trans, &error);
316 hammer_lock_ex(&nnode->lock);
317 hammer_modify_node_noundo(cursor->trans, nnode);
318 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
322 * We are not the root of the B-Tree
324 hammer_modify_node(cursor->trans, cursor->parent,
325 &elm->internal.subtree_offset,
326 sizeof(elm->internal.subtree_offset));
327 elm->internal.subtree_offset = nnode->node_offset;
328 hammer_modify_node_done(cursor->parent);
331 * We are the root of the B-Tree
333 hammer_volume_t volume;
335 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
336 KKASSERT(error == 0);
338 hammer_modify_volume_field(cursor->trans, volume,
340 volume->ondisk->vol0_btree_root = nnode->node_offset;
341 hammer_modify_volume_done(volume);
342 hammer_rel_volume(volume, 0);
345 hammer_delete_node(cursor->trans, onode);
347 if (hammer_debug_general & 0x4000) {
348 kprintf("REBLOCK LNODE %016llx -> %016llx\n",
349 onode->node_offset, nnode->node_offset);
351 hammer_modify_node_done(nnode);
352 cursor->node = nnode;
354 hammer_unlock(&onode->lock);
355 hammer_rel_node(onode);
361 * Reblock a B-Tree internal node. The parent must be adjusted to point to
362 * the new copy of the internal node, and the node's children's parent
363 * pointers must also be adjusted to point to the new copy.
365 * elm is a pointer to the parent element pointing at cursor.node.
368 hammer_reblock_int_node(struct hammer_ioc_reblock *reblock,
369 hammer_cursor_t cursor, hammer_btree_elm_t elm)
371 hammer_node_locklist_t locklist = NULL;
377 error = hammer_btree_lock_children(cursor, &locklist);
381 onode = cursor->node;
382 nnode = hammer_alloc_btree(cursor->trans, &error);
388 * Move the node. Adjust the parent's pointer to us first.
390 hammer_lock_ex(&nnode->lock);
391 hammer_modify_node_noundo(cursor->trans, nnode);
392 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
396 * We are not the root of the B-Tree
398 hammer_modify_node(cursor->trans, cursor->parent,
399 &elm->internal.subtree_offset,
400 sizeof(elm->internal.subtree_offset));
401 elm->internal.subtree_offset = nnode->node_offset;
402 hammer_modify_node_done(cursor->parent);
405 * We are the root of the B-Tree
407 hammer_volume_t volume;
409 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
410 KKASSERT(error == 0);
412 hammer_modify_volume_field(cursor->trans, volume,
414 volume->ondisk->vol0_btree_root = nnode->node_offset;
415 hammer_modify_volume_done(volume);
416 hammer_rel_volume(volume, 0);
420 * Now adjust our children's pointers to us.
422 for (i = 0; i < nnode->ondisk->count; ++i) {
423 elm = &nnode->ondisk->elms[i];
424 error = btree_set_parent(cursor->trans, nnode, elm);
426 panic("reblock internal node: fixup problem");
432 * The new node replaces the current node in the cursor. The cursor
433 * expects it to be locked so leave it locked. Discard onode.
435 hammer_delete_node(cursor->trans, onode);
437 if (hammer_debug_general & 0x4000) {
438 kprintf("REBLOCK INODE %016llx -> %016llx\n",
439 onode->node_offset, nnode->node_offset);
441 hammer_modify_node_done(nnode);
442 cursor->node = nnode;
444 hammer_unlock(&onode->lock);
445 hammer_rel_node(onode);
448 hammer_btree_unlock_children(&locklist);