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 1.23 2008/07/03 04:24:51 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
;
67 if ((reblock
->key_beg
.localization
| reblock
->key_end
.localization
) &
68 HAMMER_LOCALIZE_PSEUDOFS_MASK
) {
71 if (reblock
->key_beg
.obj_id
>= reblock
->key_end
.obj_id
)
73 if (reblock
->free_level
< 0)
76 reblock
->key_cur
= reblock
->key_beg
;
77 reblock
->key_cur
.localization
+= ip
->obj_localization
;
81 error
= hammer_init_cursor(trans
, &cursor
, NULL
, NULL
);
83 hammer_done_cursor(&cursor
);
86 cursor
.key_beg
.localization
= reblock
->key_cur
.localization
;
87 cursor
.key_beg
.obj_id
= reblock
->key_cur
.obj_id
;
88 cursor
.key_beg
.key
= HAMMER_MIN_KEY
;
89 cursor
.key_beg
.create_tid
= 1;
90 cursor
.key_beg
.delete_tid
= 0;
91 cursor
.key_beg
.rec_type
= HAMMER_MIN_RECTYPE
;
92 cursor
.key_beg
.obj_type
= 0;
94 cursor
.key_end
.localization
= reblock
->key_end
.localization
+
96 cursor
.key_end
.obj_id
= reblock
->key_end
.obj_id
;
97 cursor
.key_end
.key
= HAMMER_MAX_KEY
;
98 cursor
.key_end
.create_tid
= HAMMER_MAX_TID
- 1;
99 cursor
.key_end
.delete_tid
= 0;
100 cursor
.key_end
.rec_type
= HAMMER_MAX_RECTYPE
;
101 cursor
.key_end
.obj_type
= 0;
103 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
;
104 cursor
.flags
|= HAMMER_CURSOR_BACKEND
;
107 * This flag allows the btree scan code to return internal nodes,
108 * so we can reblock them in addition to the leafs. Only specify it
109 * if we intend to reblock B-Tree nodes.
111 if (reblock
->head
.flags
& HAMMER_IOC_DO_BTREE
)
112 cursor
.flags
|= HAMMER_CURSOR_REBLOCKING
;
114 error
= hammer_btree_first(&cursor
);
117 * Internal or Leaf node
119 elm
= &cursor
.node
->ondisk
->elms
[cursor
.index
];
120 reblock
->key_cur
.obj_id
= elm
->base
.obj_id
;
121 reblock
->key_cur
.localization
= elm
->base
.localization
;
124 * Yield to more important tasks
126 if ((error
= hammer_signal_check(trans
->hmp
)) != 0)
128 if (trans
->hmp
->sync_lock
.wanted
) {
129 tsleep(trans
, 0, "hmrslo", hz
/ 10);
133 * If we build up too much meta-data we have to wait for
136 if (hammer_flusher_meta_limit(trans
->hmp
) ||
137 hammer_flusher_undo_exhausted(trans
, 2)) {
143 * If there is insufficient free space it may be due to
144 * reserved bigblocks, which flushing might fix.
146 if (hammer_checkspace(trans
->hmp
, HAMMER_CHECKSPACE_SLOP_REBLOCK
)) {
147 if (++checkspace_count
== 10) {
156 * Acquiring the sync_lock prevents the operation from
157 * crossing a synchronization boundary.
159 * NOTE: cursor.node may have changed on return.
161 hammer_sync_lock_sh(trans
);
162 error
= hammer_reblock_helper(reblock
, &cursor
, elm
);
163 hammer_sync_unlock(trans
);
165 cursor
.flags
|= HAMMER_CURSOR_ATEDISK
;
166 error
= hammer_btree_iterate(&cursor
);
172 hammer_done_cursor(&cursor
);
173 if (error
== EWOULDBLOCK
) {
174 hammer_flusher_sync(trans
->hmp
);
177 if (error
== EDEADLK
)
179 if (error
== EINTR
) {
180 reblock
->head
.flags
|= HAMMER_IOC_HEAD_INTR
;
184 reblock
->key_cur
.localization
&= HAMMER_LOCALIZE_MASK
;
189 * Reblock the B-Tree (leaf) node, record, and/or data if necessary.
191 * XXX We have no visibility into internal B-Tree nodes at the moment,
195 hammer_reblock_helper(struct hammer_ioc_reblock
*reblock
,
196 hammer_cursor_t cursor
, hammer_btree_elm_t elm
)
199 hammer_off_t tmp_offset
;
206 hmp
= cursor
->trans
->hmp
;
209 * Reblock data. Note that data embedded in a record is reblocked
210 * by the record reblock code. Data processing only occurs at leaf
211 * nodes and for RECORD element types.
213 if (cursor
->node
->ondisk
->type
!= HAMMER_BTREE_TYPE_LEAF
)
215 if (elm
->leaf
.base
.btype
!= HAMMER_BTREE_TYPE_RECORD
)
217 tmp_offset
= elm
->leaf
.data_offset
;
224 * NOTE: Localization restrictions may also have been set-up, we can't
225 * just set the match flags willy-nilly here.
227 switch(elm
->leaf
.base
.rec_type
) {
228 case HAMMER_RECTYPE_INODE
:
229 iocflags
= HAMMER_IOC_DO_INODES
;
231 case HAMMER_RECTYPE_EXT
:
232 case HAMMER_RECTYPE_FIX
:
233 case HAMMER_RECTYPE_DIRENTRY
:
234 iocflags
= HAMMER_IOC_DO_DIRS
;
236 case HAMMER_RECTYPE_DATA
:
237 case HAMMER_RECTYPE_DB
:
238 iocflags
= HAMMER_IOC_DO_DATA
;
244 if (reblock
->head
.flags
& iocflags
) {
245 ++reblock
->data_count
;
246 reblock
->data_byte_count
+= elm
->leaf
.data_len
;
247 bytes
= hammer_blockmap_getfree(hmp
, tmp_offset
, &cur
, &error
);
248 if (hammer_debug_general
& 0x4000)
249 kprintf("D %6d/%d\n", bytes
, reblock
->free_level
);
250 if (error
== 0 && (cur
== 0 || reblock
->free_level
== 0) &&
251 bytes
>= reblock
->free_level
) {
252 hammer_io_direct_uncache(hmp
, &elm
->leaf
);
253 error
= hammer_cursor_upgrade(cursor
);
255 error
= hammer_reblock_data(reblock
,
259 ++reblock
->data_moves
;
260 reblock
->data_byte_moves
+= elm
->leaf
.data_len
;
267 * Reblock a B-Tree internal or leaf node.
269 tmp_offset
= cursor
->node
->node_offset
;
270 if (cursor
->index
== 0 &&
271 error
== 0 && (reblock
->head
.flags
& HAMMER_IOC_DO_BTREE
)) {
272 ++reblock
->btree_count
;
273 bytes
= hammer_blockmap_getfree(hmp
, tmp_offset
, &cur
, &error
);
274 if (hammer_debug_general
& 0x4000)
275 kprintf("B %6d/%d\n", bytes
, reblock
->free_level
);
276 if (error
== 0 && (cur
== 0 || reblock
->free_level
== 0) &&
277 bytes
>= reblock
->free_level
) {
278 error
= hammer_cursor_upgrade(cursor
);
281 elm
= &cursor
->parent
->ondisk
->elms
[cursor
->parent_index
];
284 switch(cursor
->node
->ondisk
->type
) {
285 case HAMMER_BTREE_TYPE_LEAF
:
286 error
= hammer_reblock_leaf_node(
287 reblock
, cursor
, elm
);
289 case HAMMER_BTREE_TYPE_INTERNAL
:
290 error
= hammer_reblock_int_node(
291 reblock
, cursor
, elm
);
294 panic("Illegal B-Tree node type");
298 ++reblock
->btree_moves
;
303 hammer_cursor_downgrade(cursor
);
308 * Reblock a record's data. Both the B-Tree element and record pointers
309 * to the data must be adjusted.
312 hammer_reblock_data(struct hammer_ioc_reblock
*reblock
,
313 hammer_cursor_t cursor
, hammer_btree_elm_t elm
)
315 struct hammer_buffer
*data_buffer
= NULL
;
316 hammer_off_t ndata_offset
;
320 error
= hammer_btree_extract(cursor
, HAMMER_CURSOR_GET_DATA
|
321 HAMMER_CURSOR_GET_LEAF
);
324 ndata
= hammer_alloc_data(cursor
->trans
, elm
->leaf
.data_len
,
325 elm
->leaf
.base
.rec_type
,
326 &ndata_offset
, &data_buffer
, &error
);
333 hammer_modify_buffer(cursor
->trans
, data_buffer
, NULL
, 0);
334 bcopy(cursor
->data
, ndata
, elm
->leaf
.data_len
);
335 hammer_modify_buffer_done(data_buffer
);
337 hammer_blockmap_free(cursor
->trans
,
338 elm
->leaf
.data_offset
, elm
->leaf
.data_len
);
340 hammer_modify_node(cursor
->trans
, cursor
->node
,
341 &elm
->leaf
.data_offset
, sizeof(hammer_off_t
));
342 elm
->leaf
.data_offset
= ndata_offset
;
343 hammer_modify_node_done(cursor
->node
);
347 hammer_rel_buffer(data_buffer
, 0);
352 * Reblock a B-Tree leaf node. The parent must be adjusted to point to
353 * the new copy of the leaf node.
355 * elm is a pointer to the parent element pointing at cursor.node.
358 hammer_reblock_leaf_node(struct hammer_ioc_reblock
*reblock
,
359 hammer_cursor_t cursor
, hammer_btree_elm_t elm
)
365 onode
= cursor
->node
;
366 nnode
= hammer_alloc_btree(cursor
->trans
, &error
);
374 hammer_lock_ex(&nnode
->lock
);
375 hammer_modify_node_noundo(cursor
->trans
, nnode
);
376 bcopy(onode
->ondisk
, nnode
->ondisk
, sizeof(*nnode
->ondisk
));
380 * We are not the root of the B-Tree
382 hammer_modify_node(cursor
->trans
, cursor
->parent
,
383 &elm
->internal
.subtree_offset
,
384 sizeof(elm
->internal
.subtree_offset
));
385 elm
->internal
.subtree_offset
= nnode
->node_offset
;
386 hammer_modify_node_done(cursor
->parent
);
389 * We are the root of the B-Tree
391 hammer_volume_t volume
;
393 volume
= hammer_get_root_volume(cursor
->trans
->hmp
, &error
);
394 KKASSERT(error
== 0);
396 hammer_modify_volume_field(cursor
->trans
, volume
,
398 volume
->ondisk
->vol0_btree_root
= nnode
->node_offset
;
399 hammer_modify_volume_done(volume
);
400 hammer_rel_volume(volume
, 0);
403 hammer_delete_node(cursor
->trans
, onode
);
405 if (hammer_debug_general
& 0x4000) {
406 kprintf("REBLOCK LNODE %016llx -> %016llx\n",
407 onode
->node_offset
, nnode
->node_offset
);
409 hammer_modify_node_done(nnode
);
410 cursor
->node
= nnode
;
412 hammer_unlock(&onode
->lock
);
413 hammer_rel_node(onode
);
419 * Reblock a B-Tree internal node. The parent must be adjusted to point to
420 * the new copy of the internal node, and the node's children's parent
421 * pointers must also be adjusted to point to the new copy.
423 * elm is a pointer to the parent element pointing at cursor.node.
426 hammer_reblock_int_node(struct hammer_ioc_reblock
*reblock
,
427 hammer_cursor_t cursor
, hammer_btree_elm_t elm
)
429 hammer_node_locklist_t locklist
= NULL
;
435 error
= hammer_btree_lock_children(cursor
, &locklist
);
439 onode
= cursor
->node
;
440 nnode
= hammer_alloc_btree(cursor
->trans
, &error
);
446 * Move the node. Adjust the parent's pointer to us first.
448 hammer_lock_ex(&nnode
->lock
);
449 hammer_modify_node_noundo(cursor
->trans
, nnode
);
450 bcopy(onode
->ondisk
, nnode
->ondisk
, sizeof(*nnode
->ondisk
));
454 * We are not the root of the B-Tree
456 hammer_modify_node(cursor
->trans
, cursor
->parent
,
457 &elm
->internal
.subtree_offset
,
458 sizeof(elm
->internal
.subtree_offset
));
459 elm
->internal
.subtree_offset
= nnode
->node_offset
;
460 hammer_modify_node_done(cursor
->parent
);
463 * We are the root of the B-Tree
465 hammer_volume_t volume
;
467 volume
= hammer_get_root_volume(cursor
->trans
->hmp
, &error
);
468 KKASSERT(error
== 0);
470 hammer_modify_volume_field(cursor
->trans
, volume
,
472 volume
->ondisk
->vol0_btree_root
= nnode
->node_offset
;
473 hammer_modify_volume_done(volume
);
474 hammer_rel_volume(volume
, 0);
478 * Now adjust our children's pointers to us.
480 for (i
= 0; i
< nnode
->ondisk
->count
; ++i
) {
481 elm
= &nnode
->ondisk
->elms
[i
];
482 error
= btree_set_parent(cursor
->trans
, nnode
, elm
);
484 panic("reblock internal node: fixup problem");
490 * The new node replaces the current node in the cursor. The cursor
491 * expects it to be locked so leave it locked. Discard onode.
493 hammer_delete_node(cursor
->trans
, onode
);
495 if (hammer_debug_general
& 0x4000) {
496 kprintf("REBLOCK INODE %016llx -> %016llx\n",
497 onode
->node_offset
, nnode
->node_offset
);
499 hammer_modify_node_done(nnode
);
500 cursor
->node
= nnode
;
502 hammer_unlock(&onode
->lock
);
503 hammer_rel_node(onode
);
506 hammer_btree_unlock_children(&locklist
);