HAMMER 38B/Many: Undo/Synchronization and crash recovery
[dragonfly.git] / sys / vfs / hammer / hammer_reblock.c
CommitLineData
bf686dbe
MD
1/*
2 * Copyright (c) 2008 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
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
16 * distribution.
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.
20 *
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
32 * SUCH DAMAGE.
33 *
6e1e8b6d 34 * $DragonFly: src/sys/vfs/hammer/hammer_reblock.c,v 1.6 2008/03/26 04:32:54 dillon Exp $
bf686dbe
MD
35 */
36/*
37 * HAMMER reblocker - This code frees up fragmented physical space
38 *
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.
42 *
43 * This function is called from an ioctl or via the hammer support thread.
44 */
45
46#include "hammer.h"
47
36f82b23 48static int hammer_reblock_helper(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
49 hammer_cursor_t cursor,
50 hammer_btree_elm_t elm);
36f82b23 51static int hammer_reblock_data(struct hammer_ioc_reblock *reblock,
bf686dbe 52 hammer_cursor_t cursor, hammer_btree_elm_t elm);
36f82b23 53static int hammer_reblock_record(struct hammer_ioc_reblock *reblock,
bf686dbe 54 hammer_cursor_t cursor, hammer_btree_elm_t elm);
36f82b23 55static int hammer_reblock_node(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
56 hammer_cursor_t cursor, hammer_btree_elm_t elm);
57
58int
36f82b23
MD
59hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
60 struct hammer_ioc_reblock *reblock)
bf686dbe
MD
61{
62 struct hammer_cursor cursor;
63 hammer_btree_elm_t elm;
64 int error;
65
66 if (reblock->beg_obj_id >= reblock->end_obj_id)
67 return(EINVAL);
68 if (reblock->free_level < 0)
69 return(EINVAL);
70
71retry:
36f82b23 72 error = hammer_init_cursor(trans, &cursor, NULL);
bf686dbe
MD
73 if (error) {
74 hammer_done_cursor(&cursor);
75 return(error);
76 }
77 cursor.key_beg.obj_id = reblock->cur_obj_id;
78 cursor.key_beg.key = HAMMER_MIN_KEY;
79 cursor.key_beg.create_tid = 1;
80 cursor.key_beg.delete_tid = 0;
81 cursor.key_beg.rec_type = HAMMER_MIN_RECTYPE;
82 cursor.key_beg.obj_type = 0;
83
84 cursor.key_end.obj_id = reblock->end_obj_id;
85 cursor.key_end.key = HAMMER_MAX_KEY;
86 cursor.key_end.create_tid = HAMMER_MAX_TID - 1;
87 cursor.key_end.delete_tid = 0;
88 cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
89 cursor.key_end.obj_type = 0;
90
91 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
92
93 error = hammer_btree_first(&cursor);
94 while (error == 0) {
95 elm = &cursor.node->ondisk->elms[cursor.index];
96 reblock->cur_obj_id = elm->base.obj_id;
97
36f82b23 98 error = hammer_reblock_helper(reblock, &cursor, elm);
bf686dbe
MD
99 if (error == 0) {
100 cursor.flags |= HAMMER_CURSOR_ATEDISK;
101 error = hammer_btree_iterate(&cursor);
102 }
855942b6
MD
103 if (error == 0)
104 error = hammer_signal_check(trans->hmp);
bf686dbe
MD
105 }
106 if (error == ENOENT)
107 error = 0;
108 hammer_done_cursor(&cursor);
109 if (error == EDEADLK)
110 goto retry;
111 return(error);
112}
113
114/*
115 * Reblock the B-Tree (leaf) node, record, and/or data if necessary.
116 *
117 * XXX We have no visibility into internal B-Tree nodes at the moment.
118 */
119static int
36f82b23 120hammer_reblock_helper(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
121 hammer_cursor_t cursor, hammer_btree_elm_t elm)
122{
123 hammer_off_t tmp_offset;
124 int error;
125 int zone;
126 int bytes;
127 int cur;
128
129 if (elm->leaf.base.btype != HAMMER_BTREE_TYPE_RECORD)
130 return(0);
131 error = 0;
132
133 /*
134 * Reblock data. Note that data embedded in a record is reblocked
135 * by the record reblock code.
136 */
137 tmp_offset = elm->leaf.data_offset;
138 zone = HAMMER_ZONE_DECODE(tmp_offset); /* can be 0 */
139 if ((zone == HAMMER_ZONE_SMALL_DATA_INDEX ||
140 zone == HAMMER_ZONE_LARGE_DATA_INDEX) && error == 0) {
141 ++reblock->data_count;
142 reblock->data_byte_count += elm->leaf.data_len;
36f82b23
MD
143 bytes = hammer_blockmap_getfree(cursor->trans->hmp, tmp_offset,
144 &cur, &error);
bf686dbe 145 if (error == 0 && cur == 0 && bytes > reblock->free_level) {
6e1e8b6d
MD
146 if (hammer_debug_general & 0x4000)
147 kprintf("%6d ", bytes);
bf686dbe
MD
148 error = hammer_cursor_upgrade(cursor);
149 if (error == 0) {
36f82b23 150 error = hammer_reblock_data(reblock,
bf686dbe
MD
151 cursor, elm);
152 }
153 if (error == 0) {
154 ++reblock->data_moves;
155 reblock->data_byte_moves += elm->leaf.data_len;
156 }
157 }
158 }
159
160 /*
161 * Reblock a record
162 */
163 tmp_offset = elm->leaf.rec_offset;
164 zone = HAMMER_ZONE_DECODE(tmp_offset);
165 if (zone == HAMMER_ZONE_RECORD_INDEX && error == 0) {
166 ++reblock->record_count;
36f82b23
MD
167 bytes = hammer_blockmap_getfree(cursor->trans->hmp, tmp_offset,
168 &cur, &error);
bf686dbe 169 if (error == 0 && cur == 0 && bytes > reblock->free_level) {
6e1e8b6d
MD
170 if (hammer_debug_general & 0x4000)
171 kprintf("%6d ", bytes);
bf686dbe
MD
172 error = hammer_cursor_upgrade(cursor);
173 if (error == 0) {
36f82b23 174 error = hammer_reblock_record(reblock,
bf686dbe
MD
175 cursor, elm);
176 }
177 if (error == 0) {
178 ++reblock->record_moves;
179 }
180 }
181 }
182
183 /*
184 * Reblock a B-Tree node. Adjust elm to point at the parent's
185 * leaf entry.
186 */
187 tmp_offset = cursor->node->node_offset;
188 zone = HAMMER_ZONE_DECODE(tmp_offset);
189 if (zone == HAMMER_ZONE_BTREE_INDEX && error == 0 &&
190 cursor->index == 0) {
191 ++reblock->btree_count;
36f82b23
MD
192 bytes = hammer_blockmap_getfree(cursor->trans->hmp, tmp_offset,
193 &cur, &error);
bf686dbe 194 if (error == 0 && cur == 0 && bytes > reblock->free_level) {
6e1e8b6d
MD
195 if (hammer_debug_general & 0x4000)
196 kprintf("%6d ", bytes);
bf686dbe
MD
197 error = hammer_cursor_upgrade(cursor);
198 if (error == 0) {
199 if (cursor->parent)
200 elm = &cursor->parent->ondisk->elms[cursor->parent_index];
201 else
202 elm = NULL;
36f82b23 203 error = hammer_reblock_node(reblock,
bf686dbe
MD
204 cursor, elm);
205 }
206 if (error == 0) {
207 ++reblock->btree_moves;
208 }
209 }
210 }
211
212 hammer_cursor_downgrade(cursor);
213 return(error);
214}
215
216/*
217 * Reblock a record's data. Both the B-Tree element and record pointers
218 * to the data must be adjusted.
219 */
220static int
36f82b23 221hammer_reblock_data(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
222 hammer_cursor_t cursor, hammer_btree_elm_t elm)
223{
224 struct hammer_buffer *data_buffer = NULL;
225 hammer_off_t ndata_offset;
226 int error;
227 void *ndata;
228
229 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA |
230 HAMMER_CURSOR_GET_RECORD);
231 if (error)
232 return (error);
36f82b23
MD
233 ndata = hammer_alloc_data(cursor->trans, elm->leaf.data_len,
234 &ndata_offset, &data_buffer, &error);
bf686dbe
MD
235 if (error)
236 goto done;
237
238 /*
239 * Move the data
240 */
241 bcopy(cursor->data, ndata, elm->leaf.data_len);
36f82b23
MD
242 hammer_modify_node(cursor->trans, cursor->node,
243 &elm->leaf.data_offset, sizeof(hammer_off_t));
244 hammer_modify_record(cursor->trans, cursor->record_buffer,
245 &cursor->record->base.data_off,
246 sizeof(hammer_off_t));
bf686dbe 247
36f82b23
MD
248 hammer_blockmap_free(cursor->trans,
249 elm->leaf.data_offset, elm->leaf.data_len);
bf686dbe 250
bf686dbe
MD
251 cursor->record->base.data_off = ndata_offset;
252 elm->leaf.data_offset = ndata_offset;
253
254done:
255 if (data_buffer)
256 hammer_rel_buffer(data_buffer, 0);
257 return (error);
258}
259
260/*
261 * Reblock a record. The B-Tree must be adjusted to point to the new record
262 * and the existing record must be physically destroyed so a FS rebuild
263 * does not see two versions of the same record.
264 */
265static int
36f82b23 266hammer_reblock_record(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
267 hammer_cursor_t cursor, hammer_btree_elm_t elm)
268{
269 struct hammer_buffer *rec_buffer = NULL;
270 hammer_off_t nrec_offset;
8fe2cd5d
MD
271 hammer_off_t ndata_offset;
272 hammer_record_ondisk_t orec;
bf686dbe
MD
273 hammer_record_ondisk_t nrec;
274 int error;
8fe2cd5d 275 int inline_data;
bf686dbe
MD
276
277 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_RECORD);
278 if (error)
279 return (error);
280
36f82b23
MD
281 nrec = hammer_alloc_record(cursor->trans, &nrec_offset,
282 elm->leaf.base.rec_type, &rec_buffer,
283 0, NULL, NULL, &error);
bf686dbe
MD
284 if (error)
285 goto done;
286
287 /*
8fe2cd5d
MD
288 * Move the record. Check for an inline data reference and move that
289 * too if necessary.
bf686dbe 290 */
8fe2cd5d
MD
291 orec = cursor->record;
292 bcopy(orec, nrec, sizeof(*nrec));
293
294 if ((orec->base.data_off & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_RECORD) {
295 ndata_offset = orec->base.data_off - elm->leaf.rec_offset;
296 KKASSERT(ndata_offset < sizeof(*nrec));
297 ndata_offset += nrec_offset;
298 inline_data = 1;
299 } else {
300 ndata_offset = 0;
301 inline_data = 0;
302 }
bf686dbe 303
36f82b23
MD
304 hammer_modify_record(cursor->trans, cursor->record_buffer,
305 &orec->base.base.rec_type,
306 sizeof(orec->base.base.rec_type));
8fe2cd5d 307 orec->base.base.rec_type |= HAMMER_RECTYPE_MOVED;
bf686dbe 308
36f82b23
MD
309 hammer_blockmap_free(cursor->trans,
310 elm->leaf.rec_offset, sizeof(*nrec));
bf686dbe 311
b58c6388
MD
312 if (hammer_debug_general & 0x4000) {
313 kprintf("REBLOCK RECD %016llx -> %016llx\n",
314 elm->leaf.rec_offset, nrec_offset);
315 }
8fe2cd5d 316
36f82b23
MD
317 hammer_modify_node(cursor->trans, cursor->node,
318 &elm->leaf.rec_offset, sizeof(hammer_off_t));
bf686dbe 319 elm->leaf.rec_offset = nrec_offset;
8fe2cd5d 320 if (inline_data) {
36f82b23
MD
321 hammer_modify_node(cursor->trans, cursor->node,
322 &elm->leaf.data_offset, sizeof(hammer_off_t));
8fe2cd5d 323 elm->leaf.data_offset = ndata_offset;
6e1e8b6d 324 nrec->base.data_off = ndata_offset;
8fe2cd5d 325 }
bf686dbe
MD
326
327done:
328 if (rec_buffer)
329 hammer_rel_buffer(rec_buffer, 0);
330 return (error);
331}
332
333/*
334 * Reblock a B-Tree (leaf) node. The parent must be adjusted to point to
335 * the new copy of the leaf node. elm is a pointer to the parent element
336 * pointing at cursor.node.
337 *
338 * XXX reblock internal nodes too.
339 */
340static int
36f82b23 341hammer_reblock_node(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
342 hammer_cursor_t cursor, hammer_btree_elm_t elm)
343{
344 hammer_node_t onode;
345 hammer_node_t nnode;
346 int error;
347
348 onode = cursor->node;
36f82b23 349 nnode = hammer_alloc_btree(cursor->trans, &error);
bf686dbe
MD
350 hammer_lock_ex(&nnode->lock);
351
352 if (nnode == NULL)
353 return (error);
354
355 /*
356 * Move the node
357 */
358 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
359
360 if (elm) {
361 /*
362 * We are not the root of the B-Tree
363 */
36f82b23 364 hammer_modify_node(cursor->trans, cursor->parent,
bf686dbe
MD
365 &elm->internal.subtree_offset,
366 sizeof(elm->internal.subtree_offset));
367 elm->internal.subtree_offset = nnode->node_offset;
368 } else {
369 /*
370 * We are the root of the B-Tree
371 */
372 hammer_volume_t volume;
373
36f82b23 374 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
bf686dbe
MD
375 KKASSERT(error == 0);
376
36f82b23
MD
377 hammer_modify_volume(cursor->trans, volume,
378 &volume->ondisk->vol0_btree_root,
bf686dbe
MD
379 sizeof(hammer_off_t));
380 volume->ondisk->vol0_btree_root = nnode->node_offset;
381 hammer_rel_volume(volume, 0);
382 }
383
36f82b23 384 hammer_delete_node(cursor->trans, onode);
bf686dbe 385
b58c6388
MD
386 if (hammer_debug_general & 0x4000) {
387 kprintf("REBLOCK NODE %016llx -> %016llx\n",
388 onode->node_offset, nnode->node_offset);
389 }
bf686dbe
MD
390
391 cursor->node = nnode;
392 hammer_rel_node(onode);
393
394 return (error);
395}
396