HAMMER 40D/Many: Inode/link-count sequencer cleanup pass.
[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 *
4e17f465 34 * $DragonFly: src/sys/vfs/hammer/hammer_reblock.c,v 1.9 2008/05/03 05:28:55 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:
4e17f465 72 error = hammer_init_cursor(trans, &cursor, NULL, 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;
9480ff55 92 cursor.flags |= HAMMER_CURSOR_BACKEND;
bf686dbe
MD
93
94 error = hammer_btree_first(&cursor);
95 while (error == 0) {
96 elm = &cursor.node->ondisk->elms[cursor.index];
97 reblock->cur_obj_id = elm->base.obj_id;
98
9480ff55
MD
99 /*
100 * Acquiring the sync_lock prevents the operation from
101 * crossing a synchronization boundary.
102 */
103 hammer_lock_ex(&trans->hmp->sync_lock);
36f82b23 104 error = hammer_reblock_helper(reblock, &cursor, elm);
9480ff55 105 hammer_unlock(&trans->hmp->sync_lock);
bf686dbe
MD
106 if (error == 0) {
107 cursor.flags |= HAMMER_CURSOR_ATEDISK;
108 error = hammer_btree_iterate(&cursor);
109 }
9480ff55
MD
110
111 /*
112 * Bad hack for now, don't blow out the kernel's buffer
113 * cache.
114 */
115 if (trans->hmp->locked_dirty_count > hammer_limit_dirtybufs)
116 hammer_flusher_sync(trans->hmp);
855942b6
MD
117 if (error == 0)
118 error = hammer_signal_check(trans->hmp);
bf686dbe
MD
119 }
120 if (error == ENOENT)
121 error = 0;
122 hammer_done_cursor(&cursor);
123 if (error == EDEADLK)
124 goto retry;
125 return(error);
126}
127
128/*
129 * Reblock the B-Tree (leaf) node, record, and/or data if necessary.
130 *
9480ff55
MD
131 * XXX We have no visibility into internal B-Tree nodes at the moment,
132 * only leaf nodes.
bf686dbe
MD
133 */
134static int
36f82b23 135hammer_reblock_helper(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
136 hammer_cursor_t cursor, hammer_btree_elm_t elm)
137{
138 hammer_off_t tmp_offset;
139 int error;
140 int zone;
141 int bytes;
142 int cur;
143
144 if (elm->leaf.base.btype != HAMMER_BTREE_TYPE_RECORD)
145 return(0);
146 error = 0;
147
148 /*
149 * Reblock data. Note that data embedded in a record is reblocked
150 * by the record reblock code.
151 */
152 tmp_offset = elm->leaf.data_offset;
153 zone = HAMMER_ZONE_DECODE(tmp_offset); /* can be 0 */
154 if ((zone == HAMMER_ZONE_SMALL_DATA_INDEX ||
155 zone == HAMMER_ZONE_LARGE_DATA_INDEX) && error == 0) {
156 ++reblock->data_count;
157 reblock->data_byte_count += elm->leaf.data_len;
36f82b23
MD
158 bytes = hammer_blockmap_getfree(cursor->trans->hmp, tmp_offset,
159 &cur, &error);
bf686dbe 160 if (error == 0 && cur == 0 && bytes > reblock->free_level) {
6e1e8b6d
MD
161 if (hammer_debug_general & 0x4000)
162 kprintf("%6d ", bytes);
bf686dbe
MD
163 error = hammer_cursor_upgrade(cursor);
164 if (error == 0) {
36f82b23 165 error = hammer_reblock_data(reblock,
bf686dbe
MD
166 cursor, elm);
167 }
168 if (error == 0) {
169 ++reblock->data_moves;
170 reblock->data_byte_moves += elm->leaf.data_len;
171 }
172 }
173 }
174
175 /*
176 * Reblock a record
177 */
178 tmp_offset = elm->leaf.rec_offset;
179 zone = HAMMER_ZONE_DECODE(tmp_offset);
180 if (zone == HAMMER_ZONE_RECORD_INDEX && error == 0) {
181 ++reblock->record_count;
36f82b23
MD
182 bytes = hammer_blockmap_getfree(cursor->trans->hmp, tmp_offset,
183 &cur, &error);
bf686dbe 184 if (error == 0 && cur == 0 && bytes > reblock->free_level) {
6e1e8b6d
MD
185 if (hammer_debug_general & 0x4000)
186 kprintf("%6d ", bytes);
bf686dbe
MD
187 error = hammer_cursor_upgrade(cursor);
188 if (error == 0) {
36f82b23 189 error = hammer_reblock_record(reblock,
bf686dbe
MD
190 cursor, elm);
191 }
192 if (error == 0) {
193 ++reblock->record_moves;
194 }
195 }
196 }
197
198 /*
199 * Reblock a B-Tree node. Adjust elm to point at the parent's
200 * leaf entry.
201 */
202 tmp_offset = cursor->node->node_offset;
203 zone = HAMMER_ZONE_DECODE(tmp_offset);
204 if (zone == HAMMER_ZONE_BTREE_INDEX && error == 0 &&
205 cursor->index == 0) {
206 ++reblock->btree_count;
36f82b23
MD
207 bytes = hammer_blockmap_getfree(cursor->trans->hmp, tmp_offset,
208 &cur, &error);
bf686dbe 209 if (error == 0 && cur == 0 && bytes > reblock->free_level) {
6e1e8b6d
MD
210 if (hammer_debug_general & 0x4000)
211 kprintf("%6d ", bytes);
bf686dbe
MD
212 error = hammer_cursor_upgrade(cursor);
213 if (error == 0) {
214 if (cursor->parent)
215 elm = &cursor->parent->ondisk->elms[cursor->parent_index];
216 else
217 elm = NULL;
36f82b23 218 error = hammer_reblock_node(reblock,
bf686dbe
MD
219 cursor, elm);
220 }
221 if (error == 0) {
222 ++reblock->btree_moves;
223 }
224 }
225 }
226
227 hammer_cursor_downgrade(cursor);
228 return(error);
229}
230
231/*
232 * Reblock a record's data. Both the B-Tree element and record pointers
233 * to the data must be adjusted.
234 */
235static int
36f82b23 236hammer_reblock_data(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
237 hammer_cursor_t cursor, hammer_btree_elm_t elm)
238{
239 struct hammer_buffer *data_buffer = NULL;
240 hammer_off_t ndata_offset;
241 int error;
242 void *ndata;
243
244 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA |
245 HAMMER_CURSOR_GET_RECORD);
246 if (error)
247 return (error);
36f82b23
MD
248 ndata = hammer_alloc_data(cursor->trans, elm->leaf.data_len,
249 &ndata_offset, &data_buffer, &error);
bf686dbe
MD
250 if (error)
251 goto done;
252
253 /*
254 * Move the data
255 */
10a5d1ba 256 hammer_modify_buffer(cursor->trans, data_buffer, NULL, 0);
bf686dbe 257 bcopy(cursor->data, ndata, elm->leaf.data_len);
10a5d1ba 258 hammer_modify_buffer_done(data_buffer);
bf686dbe 259
36f82b23
MD
260 hammer_blockmap_free(cursor->trans,
261 elm->leaf.data_offset, elm->leaf.data_len);
bf686dbe 262
10a5d1ba
MD
263 hammer_modify_record(cursor->trans, cursor->record_buffer,
264 &cursor->record->base.data_off,
265 sizeof(hammer_off_t));
bf686dbe 266 cursor->record->base.data_off = ndata_offset;
10a5d1ba
MD
267 hammer_modify_record_done(cursor->record_buffer);
268
269 hammer_modify_node(cursor->trans, cursor->node,
270 &elm->leaf.data_offset, sizeof(hammer_off_t));
bf686dbe 271 elm->leaf.data_offset = ndata_offset;
10a5d1ba 272 hammer_modify_node_done(cursor->node);
bf686dbe
MD
273
274done:
275 if (data_buffer)
276 hammer_rel_buffer(data_buffer, 0);
277 return (error);
278}
279
280/*
281 * Reblock a record. The B-Tree must be adjusted to point to the new record
282 * and the existing record must be physically destroyed so a FS rebuild
283 * does not see two versions of the same record.
284 */
285static int
36f82b23 286hammer_reblock_record(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
287 hammer_cursor_t cursor, hammer_btree_elm_t elm)
288{
289 struct hammer_buffer *rec_buffer = NULL;
290 hammer_off_t nrec_offset;
8fe2cd5d
MD
291 hammer_off_t ndata_offset;
292 hammer_record_ondisk_t orec;
bf686dbe
MD
293 hammer_record_ondisk_t nrec;
294 int error;
8fe2cd5d 295 int inline_data;
bf686dbe
MD
296
297 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_RECORD);
298 if (error)
299 return (error);
300
36f82b23
MD
301 nrec = hammer_alloc_record(cursor->trans, &nrec_offset,
302 elm->leaf.base.rec_type, &rec_buffer,
303 0, NULL, NULL, &error);
bf686dbe
MD
304 if (error)
305 goto done;
306
307 /*
8fe2cd5d
MD
308 * Move the record. Check for an inline data reference and move that
309 * too if necessary.
bf686dbe 310 */
8fe2cd5d 311 orec = cursor->record;
10a5d1ba 312 hammer_modify_buffer(cursor->trans, rec_buffer, NULL, 0);
8fe2cd5d
MD
313 bcopy(orec, nrec, sizeof(*nrec));
314
315 if ((orec->base.data_off & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_RECORD) {
316 ndata_offset = orec->base.data_off - elm->leaf.rec_offset;
317 KKASSERT(ndata_offset < sizeof(*nrec));
318 ndata_offset += nrec_offset;
319 inline_data = 1;
320 } else {
321 ndata_offset = 0;
322 inline_data = 0;
323 }
bf686dbe 324
36f82b23
MD
325 hammer_modify_record(cursor->trans, cursor->record_buffer,
326 &orec->base.base.rec_type,
327 sizeof(orec->base.base.rec_type));
8fe2cd5d 328 orec->base.base.rec_type |= HAMMER_RECTYPE_MOVED;
10a5d1ba 329 hammer_modify_record_done(cursor->record_buffer);
bf686dbe 330
36f82b23
MD
331 hammer_blockmap_free(cursor->trans,
332 elm->leaf.rec_offset, sizeof(*nrec));
bf686dbe 333
b58c6388
MD
334 if (hammer_debug_general & 0x4000) {
335 kprintf("REBLOCK RECD %016llx -> %016llx\n",
336 elm->leaf.rec_offset, nrec_offset);
337 }
8fe2cd5d 338
36f82b23
MD
339 hammer_modify_node(cursor->trans, cursor->node,
340 &elm->leaf.rec_offset, sizeof(hammer_off_t));
bf686dbe 341 elm->leaf.rec_offset = nrec_offset;
10a5d1ba 342 hammer_modify_node_done(cursor->node);
8fe2cd5d 343 if (inline_data) {
36f82b23
MD
344 hammer_modify_node(cursor->trans, cursor->node,
345 &elm->leaf.data_offset, sizeof(hammer_off_t));
8fe2cd5d 346 elm->leaf.data_offset = ndata_offset;
10a5d1ba 347 hammer_modify_node_done(cursor->node);
6e1e8b6d 348 nrec->base.data_off = ndata_offset;
8fe2cd5d 349 }
10a5d1ba 350 hammer_modify_buffer_done(rec_buffer);
bf686dbe
MD
351
352done:
353 if (rec_buffer)
354 hammer_rel_buffer(rec_buffer, 0);
355 return (error);
356}
357
358/*
359 * Reblock a B-Tree (leaf) node. The parent must be adjusted to point to
360 * the new copy of the leaf node. elm is a pointer to the parent element
361 * pointing at cursor.node.
362 *
363 * XXX reblock internal nodes too.
364 */
365static int
36f82b23 366hammer_reblock_node(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
367 hammer_cursor_t cursor, hammer_btree_elm_t elm)
368{
369 hammer_node_t onode;
370 hammer_node_t nnode;
371 int error;
372
373 onode = cursor->node;
36f82b23 374 nnode = hammer_alloc_btree(cursor->trans, &error);
bf686dbe
MD
375 hammer_lock_ex(&nnode->lock);
376
377 if (nnode == NULL)
378 return (error);
379
380 /*
381 * Move the node
382 */
383 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
384
385 if (elm) {
386 /*
387 * We are not the root of the B-Tree
388 */
36f82b23 389 hammer_modify_node(cursor->trans, cursor->parent,
bf686dbe
MD
390 &elm->internal.subtree_offset,
391 sizeof(elm->internal.subtree_offset));
392 elm->internal.subtree_offset = nnode->node_offset;
10a5d1ba 393 hammer_modify_node_done(cursor->parent);
bf686dbe
MD
394 } else {
395 /*
396 * We are the root of the B-Tree
397 */
398 hammer_volume_t volume;
399
36f82b23 400 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
bf686dbe
MD
401 KKASSERT(error == 0);
402
36f82b23
MD
403 hammer_modify_volume(cursor->trans, volume,
404 &volume->ondisk->vol0_btree_root,
bf686dbe
MD
405 sizeof(hammer_off_t));
406 volume->ondisk->vol0_btree_root = nnode->node_offset;
10a5d1ba 407 hammer_modify_volume_done(volume);
bf686dbe
MD
408 hammer_rel_volume(volume, 0);
409 }
410
36f82b23 411 hammer_delete_node(cursor->trans, onode);
bf686dbe 412
b58c6388
MD
413 if (hammer_debug_general & 0x4000) {
414 kprintf("REBLOCK NODE %016llx -> %016llx\n",
415 onode->node_offset, nnode->node_offset);
416 }
bf686dbe
MD
417
418 cursor->node = nnode;
419 hammer_rel_node(onode);
420
421 return (error);
422}
423