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 1.29 2008/07/12 23:04:50 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;
68 if ((reblock->key_beg.localization | reblock->key_end.localization) &
69 HAMMER_LOCALIZE_PSEUDOFS_MASK) {
72 if (reblock->key_beg.obj_id >= reblock->key_end.obj_id)
74 if (reblock->free_level < 0)
77 reblock->key_cur = reblock->key_beg;
78 reblock->key_cur.localization &= HAMMER_LOCALIZE_MASK;
79 reblock->key_cur.localization += ip->obj_localization;
82 seq = trans->hmp->flusher.act;
84 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
86 hammer_done_cursor(&cursor);
89 cursor.key_beg.localization = reblock->key_cur.localization;
90 cursor.key_beg.obj_id = reblock->key_cur.obj_id;
91 cursor.key_beg.key = HAMMER_MIN_KEY;
92 cursor.key_beg.create_tid = 1;
93 cursor.key_beg.delete_tid = 0;
94 cursor.key_beg.rec_type = HAMMER_MIN_RECTYPE;
95 cursor.key_beg.obj_type = 0;
97 cursor.key_end.localization = (reblock->key_end.localization &
98 HAMMER_LOCALIZE_MASK) +
100 cursor.key_end.obj_id = reblock->key_end.obj_id;
101 cursor.key_end.key = HAMMER_MAX_KEY;
102 cursor.key_end.create_tid = HAMMER_MAX_TID - 1;
103 cursor.key_end.delete_tid = 0;
104 cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
105 cursor.key_end.obj_type = 0;
107 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
108 cursor.flags |= HAMMER_CURSOR_BACKEND;
111 * This flag allows the btree scan code to return internal nodes,
112 * so we can reblock them in addition to the leafs. Only specify it
113 * if we intend to reblock B-Tree nodes.
115 if (reblock->head.flags & HAMMER_IOC_DO_BTREE)
116 cursor.flags |= HAMMER_CURSOR_REBLOCKING;
118 error = hammer_btree_first(&cursor);
121 * Internal or Leaf node
123 elm = &cursor.node->ondisk->elms[cursor.index];
124 reblock->key_cur.obj_id = elm->base.obj_id;
125 reblock->key_cur.localization = elm->base.localization;
128 * Yield to more important tasks
130 if ((error = hammer_signal_check(trans->hmp)) != 0)
134 * If there is insufficient free space it may be due to
135 * reserved bigblocks, which flushing might fix.
137 if (hammer_checkspace(trans->hmp, HAMMER_CHKSPC_REBLOCK)) {
138 if (++checkspace_count == 10) {
142 hammer_unlock_cursor(&cursor, 0);
143 hammer_flusher_wait(trans->hmp, seq);
144 hammer_lock_cursor(&cursor, 0);
145 seq = hammer_flusher_async(trans->hmp, NULL);
149 * Acquiring the sync_lock prevents the operation from
150 * crossing a synchronization boundary.
152 * NOTE: cursor.node may have changed on return.
154 hammer_sync_lock_sh(trans);
155 error = hammer_reblock_helper(reblock, &cursor, elm);
156 hammer_sync_unlock(trans);
158 if (hammer_flusher_meta_halflimit(trans->hmp) ||
159 hammer_flusher_undo_exhausted(trans, 2)) {
160 hammer_unlock_cursor(&cursor, 0);
161 hammer_flusher_wait(trans->hmp, seq);
162 hammer_lock_cursor(&cursor, 0);
163 seq = hammer_flusher_async(trans->hmp, NULL);
166 cursor.flags |= HAMMER_CURSOR_ATEDISK;
167 error = hammer_btree_iterate(&cursor);
173 hammer_done_cursor(&cursor);
174 if (error == EWOULDBLOCK) {
175 hammer_flusher_sync(trans->hmp);
178 if (error == EDEADLK)
180 if (error == EINTR) {
181 reblock->head.flags |= HAMMER_IOC_HEAD_INTR;
185 reblock->key_cur.localization &= HAMMER_LOCALIZE_MASK;
190 * Reblock the B-Tree (leaf) node, record, and/or data if necessary.
192 * XXX We have no visibility into internal B-Tree nodes at the moment,
196 hammer_reblock_helper(struct hammer_ioc_reblock *reblock,
197 hammer_cursor_t cursor, hammer_btree_elm_t elm)
200 hammer_off_t tmp_offset;
207 hmp = cursor->trans->hmp;
210 * Reblock data. Note that data embedded in a record is reblocked
211 * by the record reblock code. Data processing only occurs at leaf
212 * nodes and for RECORD element types.
214 if (cursor->node->ondisk->type != HAMMER_BTREE_TYPE_LEAF)
216 if (elm->leaf.base.btype != HAMMER_BTREE_TYPE_RECORD)
218 tmp_offset = elm->leaf.data_offset;
225 * NOTE: Localization restrictions may also have been set-up, we can't
226 * just set the match flags willy-nilly here.
228 switch(elm->leaf.base.rec_type) {
229 case HAMMER_RECTYPE_INODE:
230 iocflags = HAMMER_IOC_DO_INODES;
232 case HAMMER_RECTYPE_EXT:
233 case HAMMER_RECTYPE_FIX:
234 case HAMMER_RECTYPE_PFS:
235 case HAMMER_RECTYPE_DIRENTRY:
236 iocflags = HAMMER_IOC_DO_DIRS;
238 case HAMMER_RECTYPE_DATA:
239 case HAMMER_RECTYPE_DB:
240 iocflags = HAMMER_IOC_DO_DATA;
246 if (reblock->head.flags & iocflags) {
247 ++reblock->data_count;
248 reblock->data_byte_count += elm->leaf.data_len;
249 bytes = hammer_blockmap_getfree(hmp, tmp_offset, &cur, &error);
250 if (hammer_debug_general & 0x4000)
251 kprintf("D %6d/%d\n", bytes, reblock->free_level);
252 if (error == 0 && (cur == 0 || reblock->free_level == 0) &&
253 bytes >= reblock->free_level) {
254 hammer_io_direct_uncache(hmp, &elm->leaf);
255 error = hammer_cursor_upgrade(cursor);
257 error = hammer_reblock_data(reblock,
261 ++reblock->data_moves;
262 reblock->data_byte_moves += elm->leaf.data_len;
269 * Reblock a B-Tree internal or leaf node.
271 tmp_offset = cursor->node->node_offset;
272 if (cursor->index == 0 &&
273 error == 0 && (reblock->head.flags & HAMMER_IOC_DO_BTREE)) {
274 ++reblock->btree_count;
275 bytes = hammer_blockmap_getfree(hmp, tmp_offset, &cur, &error);
276 if (hammer_debug_general & 0x4000)
277 kprintf("B %6d/%d\n", bytes, reblock->free_level);
278 if (error == 0 && (cur == 0 || reblock->free_level == 0) &&
279 bytes >= reblock->free_level) {
280 error = hammer_cursor_upgrade(cursor);
283 elm = &cursor->parent->ondisk->elms[cursor->parent_index];
286 switch(cursor->node->ondisk->type) {
287 case HAMMER_BTREE_TYPE_LEAF:
288 error = hammer_reblock_leaf_node(
289 reblock, cursor, elm);
291 case HAMMER_BTREE_TYPE_INTERNAL:
292 error = hammer_reblock_int_node(
293 reblock, cursor, elm);
296 panic("Illegal B-Tree node type");
300 ++reblock->btree_moves;
305 hammer_cursor_downgrade(cursor);
310 * Reblock a record's data. Both the B-Tree element and record pointers
311 * to the data must be adjusted.
314 hammer_reblock_data(struct hammer_ioc_reblock *reblock,
315 hammer_cursor_t cursor, hammer_btree_elm_t elm)
317 struct hammer_buffer *data_buffer = NULL;
318 hammer_off_t ndata_offset;
322 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA |
323 HAMMER_CURSOR_GET_LEAF);
326 ndata = hammer_alloc_data(cursor->trans, elm->leaf.data_len,
327 elm->leaf.base.rec_type,
328 &ndata_offset, &data_buffer, &error);
335 hammer_modify_buffer(cursor->trans, data_buffer, NULL, 0);
336 bcopy(cursor->data, ndata, elm->leaf.data_len);
337 hammer_modify_buffer_done(data_buffer);
339 hammer_blockmap_free(cursor->trans,
340 elm->leaf.data_offset, elm->leaf.data_len);
342 hammer_modify_node(cursor->trans, cursor->node,
343 &elm->leaf.data_offset, sizeof(hammer_off_t));
344 elm->leaf.data_offset = ndata_offset;
345 hammer_modify_node_done(cursor->node);
349 hammer_rel_buffer(data_buffer, 0);
354 * Reblock a B-Tree leaf node. The parent must be adjusted to point to
355 * the new copy of the leaf node.
357 * elm is a pointer to the parent element pointing at cursor.node.
360 hammer_reblock_leaf_node(struct hammer_ioc_reblock *reblock,
361 hammer_cursor_t cursor, hammer_btree_elm_t elm)
367 onode = cursor->node;
368 nnode = hammer_alloc_btree(cursor->trans, &error);
376 hammer_lock_ex(&nnode->lock);
377 hammer_modify_node_noundo(cursor->trans, nnode);
378 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
382 * We are not the root of the B-Tree
384 hammer_modify_node(cursor->trans, cursor->parent,
385 &elm->internal.subtree_offset,
386 sizeof(elm->internal.subtree_offset));
387 elm->internal.subtree_offset = nnode->node_offset;
388 hammer_modify_node_done(cursor->parent);
391 * We are the root of the B-Tree
393 hammer_volume_t volume;
395 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
396 KKASSERT(error == 0);
398 hammer_modify_volume_field(cursor->trans, volume,
400 volume->ondisk->vol0_btree_root = nnode->node_offset;
401 hammer_modify_volume_done(volume);
402 hammer_rel_volume(volume, 0);
405 hammer_cursor_replaced_node(onode, nnode);
406 hammer_delete_node(cursor->trans, onode);
408 if (hammer_debug_general & 0x4000) {
409 kprintf("REBLOCK LNODE %016llx -> %016llx\n",
410 onode->node_offset, nnode->node_offset);
412 hammer_modify_node_done(nnode);
413 cursor->node = nnode;
415 hammer_unlock(&onode->lock);
416 hammer_rel_node(onode);
422 * Reblock a B-Tree internal node. The parent must be adjusted to point to
423 * the new copy of the internal node, and the node's children's parent
424 * pointers must also be adjusted to point to the new copy.
426 * elm is a pointer to the parent element pointing at cursor.node.
429 hammer_reblock_int_node(struct hammer_ioc_reblock *reblock,
430 hammer_cursor_t cursor, hammer_btree_elm_t elm)
432 hammer_node_locklist_t locklist = NULL;
438 error = hammer_btree_lock_children(cursor, &locklist);
442 onode = cursor->node;
443 nnode = hammer_alloc_btree(cursor->trans, &error);
449 * Move the node. Adjust the parent's pointer to us first.
451 hammer_lock_ex(&nnode->lock);
452 hammer_modify_node_noundo(cursor->trans, nnode);
453 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
457 * We are not the root of the B-Tree
459 hammer_modify_node(cursor->trans, cursor->parent,
460 &elm->internal.subtree_offset,
461 sizeof(elm->internal.subtree_offset));
462 elm->internal.subtree_offset = nnode->node_offset;
463 hammer_modify_node_done(cursor->parent);
466 * We are the root of the B-Tree
468 hammer_volume_t volume;
470 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
471 KKASSERT(error == 0);
473 hammer_modify_volume_field(cursor->trans, volume,
475 volume->ondisk->vol0_btree_root = nnode->node_offset;
476 hammer_modify_volume_done(volume);
477 hammer_rel_volume(volume, 0);
481 * Now adjust our children's pointers to us.
483 for (i = 0; i < nnode->ondisk->count; ++i) {
484 elm = &nnode->ondisk->elms[i];
485 error = btree_set_parent(cursor->trans, nnode, elm);
487 panic("reblock internal node: fixup problem");
493 * The new node replaces the current node in the cursor. The cursor
494 * expects it to be locked so leave it locked. Discard onode.
496 hammer_cursor_replaced_node(onode, nnode);
497 hammer_delete_node(cursor->trans, onode);
499 if (hammer_debug_general & 0x4000) {
500 kprintf("REBLOCK INODE %016llx -> %016llx\n",
501 onode->node_offset, nnode->node_offset);
503 hammer_modify_node_done(nnode);
504 cursor->node = nnode;
506 hammer_unlock(&onode->lock);
507 hammer_rel_node(onode);
510 hammer_btree_unlock_children(&locklist);