HAMMER 56A/Many: Performance tuning - MEDIA STRUCTURES CHANGED!
[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 *
bf3b416b 34 * $DragonFly: src/sys/vfs/hammer/hammer_reblock.c,v 1.18 2008/06/14 01:42:13 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);
2f85fa4d
MD
53static int hammer_reblock_leaf_node(struct hammer_ioc_reblock *reblock,
54 hammer_cursor_t cursor, hammer_btree_elm_t elm);
55static int hammer_reblock_int_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
814387f6 71 reblock->cur_obj_id = reblock->beg_obj_id;
2f85fa4d 72 reblock->cur_localization = reblock->beg_localization;
814387f6 73
bf686dbe 74retry:
4e17f465 75 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
bf686dbe
MD
76 if (error) {
77 hammer_done_cursor(&cursor);
78 return(error);
79 }
2f85fa4d 80 cursor.key_beg.localization = reblock->cur_localization;
bf686dbe
MD
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;
87
2f85fa4d 88 cursor.key_end.localization = reblock->end_localization;
bf686dbe
MD
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;
95
96 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
9480ff55 97 cursor.flags |= HAMMER_CURSOR_BACKEND;
bf686dbe 98
2f85fa4d
MD
99 /*
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.
103 */
104 if (reblock->head.flags & HAMMER_IOC_DO_BTREE)
105 cursor.flags |= HAMMER_CURSOR_REBLOCKING;
106
bf686dbe
MD
107 error = hammer_btree_first(&cursor);
108 while (error == 0) {
2f85fa4d
MD
109 /*
110 * Internal or Leaf node
111 */
bf686dbe
MD
112 elm = &cursor.node->ondisk->elms[cursor.index];
113 reblock->cur_obj_id = elm->base.obj_id;
2f85fa4d 114 reblock->cur_localization = elm->base.localization;
bf686dbe 115
9f5097dc
MD
116 /*
117 * Yield to more important tasks
118 */
119 if ((error = hammer_signal_check(trans->hmp)) != 0)
120 break;
121 if (trans->hmp->sync_lock.wanted) {
122 tsleep(trans, 0, "hmrslo", hz / 10);
123 }
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);
128 }
129
9480ff55
MD
130 /*
131 * Acquiring the sync_lock prevents the operation from
132 * crossing a synchronization boundary.
09ac686b
MD
133 *
134 * NOTE: cursor.node may have changed on return.
9480ff55 135 */
2f85fa4d 136 hammer_sync_lock_sh(trans);
36f82b23 137 error = hammer_reblock_helper(reblock, &cursor, elm);
2f85fa4d 138 hammer_sync_unlock(trans);
bf686dbe
MD
139 if (error == 0) {
140 cursor.flags |= HAMMER_CURSOR_ATEDISK;
141 error = hammer_btree_iterate(&cursor);
142 }
143 }
144 if (error == ENOENT)
145 error = 0;
146 hammer_done_cursor(&cursor);
147 if (error == EDEADLK)
148 goto retry;
19619882
MD
149 if (error == EINTR) {
150 reblock->head.flags |= HAMMER_IOC_HEAD_INTR;
151 error = 0;
152 }
bf686dbe
MD
153 return(error);
154}
155
156/*
157 * Reblock the B-Tree (leaf) node, record, and/or data if necessary.
158 *
9480ff55
MD
159 * XXX We have no visibility into internal B-Tree nodes at the moment,
160 * only leaf nodes.
bf686dbe
MD
161 */
162static int
36f82b23 163hammer_reblock_helper(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
164 hammer_cursor_t cursor, hammer_btree_elm_t elm)
165{
166 hammer_off_t tmp_offset;
167 int error;
bf686dbe
MD
168 int bytes;
169 int cur;
bf3b416b 170 int iocflags;
bf686dbe 171
bf686dbe
MD
172 error = 0;
173
174 /*
175 * Reblock data. Note that data embedded in a record is reblocked
2f85fa4d
MD
176 * by the record reblock code. Data processing only occurs at leaf
177 * nodes and for RECORD element types.
bf686dbe 178 */
2f85fa4d
MD
179 if (cursor->node->ondisk->type != HAMMER_BTREE_TYPE_LEAF)
180 goto skip;
181 if (elm->leaf.base.btype != HAMMER_BTREE_TYPE_RECORD)
182 return(0);
bf686dbe 183 tmp_offset = elm->leaf.data_offset;
bf3b416b
MD
184 if (tmp_offset == 0)
185 goto skip;
186 if (error)
187 goto skip;
188
189 /*
190 * NOTE: Localization restrictions may also have been set-up, we can't
191 * just set the match flags willy-nilly here.
192 */
193 switch(elm->leaf.base.rec_type) {
194 case HAMMER_RECTYPE_INODE:
195 case HAMMER_RECTYPE_PSEUDO_INODE:
196 iocflags = HAMMER_IOC_DO_INODES;
197 break;
198 case HAMMER_RECTYPE_EXT:
199 case HAMMER_RECTYPE_FIX:
200 case HAMMER_RECTYPE_DIRENTRY:
201 iocflags = HAMMER_IOC_DO_DIRS;
202 break;
203 case HAMMER_RECTYPE_DATA:
204 case HAMMER_RECTYPE_DB:
205 iocflags = HAMMER_IOC_DO_DATA;
206 break;
207 default:
208 iocflags = 0;
209 break;
210 }
211 if (reblock->head.flags & iocflags) {
bf686dbe
MD
212 ++reblock->data_count;
213 reblock->data_byte_count += elm->leaf.data_len;
36f82b23
MD
214 bytes = hammer_blockmap_getfree(cursor->trans->hmp, tmp_offset,
215 &cur, &error);
2f85fa4d
MD
216 if (hammer_debug_general & 0x4000)
217 kprintf("D %6d/%d\n", bytes, reblock->free_level);
bf3b416b
MD
218 if (error == 0 && (cur == 0 || reblock->free_level == 0) &&
219 bytes >= reblock->free_level) {
bf686dbe
MD
220 error = hammer_cursor_upgrade(cursor);
221 if (error == 0) {
36f82b23 222 error = hammer_reblock_data(reblock,
bf686dbe
MD
223 cursor, elm);
224 }
225 if (error == 0) {
226 ++reblock->data_moves;
227 reblock->data_byte_moves += elm->leaf.data_len;
228 }
229 }
230 }
231
2f85fa4d 232skip:
bf686dbe 233 /*
2f85fa4d 234 * Reblock a B-Tree internal or leaf node.
bf686dbe
MD
235 */
236 tmp_offset = cursor->node->node_offset;
bf3b416b 237 if (cursor->index == 0 &&
814387f6 238 error == 0 && (reblock->head.flags & HAMMER_IOC_DO_BTREE)) {
bf686dbe 239 ++reblock->btree_count;
36f82b23
MD
240 bytes = hammer_blockmap_getfree(cursor->trans->hmp, tmp_offset,
241 &cur, &error);
2f85fa4d
MD
242 if (hammer_debug_general & 0x4000)
243 kprintf("B %6d/%d\n", bytes, reblock->free_level);
bf3b416b
MD
244 if (error == 0 && (cur == 0 || reblock->free_level == 0) &&
245 bytes >= reblock->free_level) {
bf686dbe
MD
246 error = hammer_cursor_upgrade(cursor);
247 if (error == 0) {
248 if (cursor->parent)
249 elm = &cursor->parent->ondisk->elms[cursor->parent_index];
250 else
251 elm = NULL;
2f85fa4d
MD
252 switch(cursor->node->ondisk->type) {
253 case HAMMER_BTREE_TYPE_LEAF:
254 error = hammer_reblock_leaf_node(
255 reblock, cursor, elm);
256 break;
257 case HAMMER_BTREE_TYPE_INTERNAL:
258 error = hammer_reblock_int_node(
259 reblock, cursor, elm);
260 break;
261 default:
262 panic("Illegal B-Tree node type");
263 }
bf686dbe
MD
264 }
265 if (error == 0) {
266 ++reblock->btree_moves;
267 }
268 }
269 }
270
271 hammer_cursor_downgrade(cursor);
272 return(error);
273}
274
275/*
276 * Reblock a record's data. Both the B-Tree element and record pointers
277 * to the data must be adjusted.
278 */
279static int
36f82b23 280hammer_reblock_data(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
281 hammer_cursor_t cursor, hammer_btree_elm_t elm)
282{
283 struct hammer_buffer *data_buffer = NULL;
284 hammer_off_t ndata_offset;
285 int error;
286 void *ndata;
287
288 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA |
11ad5ade 289 HAMMER_CURSOR_GET_LEAF);
bf686dbe
MD
290 if (error)
291 return (error);
36f82b23 292 ndata = hammer_alloc_data(cursor->trans, elm->leaf.data_len,
bf3b416b 293 elm->leaf.base.rec_type,
36f82b23 294 &ndata_offset, &data_buffer, &error);
bf686dbe
MD
295 if (error)
296 goto done;
297
298 /*
299 * Move the data
300 */
10a5d1ba 301 hammer_modify_buffer(cursor->trans, data_buffer, NULL, 0);
bf686dbe 302 bcopy(cursor->data, ndata, elm->leaf.data_len);
10a5d1ba 303 hammer_modify_buffer_done(data_buffer);
bf686dbe 304
36f82b23
MD
305 hammer_blockmap_free(cursor->trans,
306 elm->leaf.data_offset, elm->leaf.data_len);
bf686dbe 307
10a5d1ba
MD
308 hammer_modify_node(cursor->trans, cursor->node,
309 &elm->leaf.data_offset, sizeof(hammer_off_t));
bf686dbe 310 elm->leaf.data_offset = ndata_offset;
10a5d1ba 311 hammer_modify_node_done(cursor->node);
bf686dbe
MD
312
313done:
314 if (data_buffer)
315 hammer_rel_buffer(data_buffer, 0);
316 return (error);
317}
318
bf686dbe 319/*
2f85fa4d
MD
320 * Reblock a B-Tree leaf node. The parent must be adjusted to point to
321 * the new copy of the leaf node.
bf686dbe 322 *
2f85fa4d 323 * elm is a pointer to the parent element pointing at cursor.node.
bf686dbe
MD
324 */
325static int
2f85fa4d
MD
326hammer_reblock_leaf_node(struct hammer_ioc_reblock *reblock,
327 hammer_cursor_t cursor, hammer_btree_elm_t elm)
bf686dbe
MD
328{
329 hammer_node_t onode;
330 hammer_node_t nnode;
331 int error;
332
333 onode = cursor->node;
36f82b23 334 nnode = hammer_alloc_btree(cursor->trans, &error);
8d0efe43 335
bf686dbe
MD
336 if (nnode == NULL)
337 return (error);
338
339 /*
340 * Move the node
341 */
09ac686b
MD
342 hammer_lock_ex(&nnode->lock);
343 hammer_modify_node_noundo(cursor->trans, nnode);
bf686dbe
MD
344 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
345
346 if (elm) {
347 /*
348 * We are not the root of the B-Tree
349 */
36f82b23 350 hammer_modify_node(cursor->trans, cursor->parent,
bf686dbe
MD
351 &elm->internal.subtree_offset,
352 sizeof(elm->internal.subtree_offset));
353 elm->internal.subtree_offset = nnode->node_offset;
10a5d1ba 354 hammer_modify_node_done(cursor->parent);
bf686dbe
MD
355 } else {
356 /*
357 * We are the root of the B-Tree
358 */
359 hammer_volume_t volume;
360
36f82b23 361 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
bf686dbe
MD
362 KKASSERT(error == 0);
363
e8599db1
MD
364 hammer_modify_volume_field(cursor->trans, volume,
365 vol0_btree_root);
bf686dbe 366 volume->ondisk->vol0_btree_root = nnode->node_offset;
10a5d1ba 367 hammer_modify_volume_done(volume);
bf686dbe
MD
368 hammer_rel_volume(volume, 0);
369 }
370
36f82b23 371 hammer_delete_node(cursor->trans, onode);
bf686dbe 372
b58c6388 373 if (hammer_debug_general & 0x4000) {
2f85fa4d
MD
374 kprintf("REBLOCK LNODE %016llx -> %016llx\n",
375 onode->node_offset, nnode->node_offset);
376 }
377 hammer_modify_node_done(nnode);
378 cursor->node = nnode;
379
380 hammer_unlock(&onode->lock);
381 hammer_rel_node(onode);
382
383 return (error);
384}
385
386/*
387 * Reblock a B-Tree internal node. The parent must be adjusted to point to
388 * the new copy of the internal node, and the node's children's parent
389 * pointers must also be adjusted to point to the new copy.
390 *
391 * elm is a pointer to the parent element pointing at cursor.node.
392 */
393static int
394hammer_reblock_int_node(struct hammer_ioc_reblock *reblock,
395 hammer_cursor_t cursor, hammer_btree_elm_t elm)
396{
397 hammer_node_locklist_t locklist = NULL;
398 hammer_node_t onode;
399 hammer_node_t nnode;
400 int error;
401 int i;
402
403 error = hammer_btree_lock_children(cursor, &locklist);
404 if (error)
405 goto done;
406
407 onode = cursor->node;
408 nnode = hammer_alloc_btree(cursor->trans, &error);
409
410 if (nnode == NULL)
411 goto done;
412
413 /*
414 * Move the node. Adjust the parent's pointer to us first.
415 */
416 hammer_lock_ex(&nnode->lock);
417 hammer_modify_node_noundo(cursor->trans, nnode);
418 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
419
420 if (elm) {
421 /*
422 * We are not the root of the B-Tree
423 */
424 hammer_modify_node(cursor->trans, cursor->parent,
425 &elm->internal.subtree_offset,
426 sizeof(elm->internal.subtree_offset));
427 elm->internal.subtree_offset = nnode->node_offset;
428 hammer_modify_node_done(cursor->parent);
429 } else {
430 /*
431 * We are the root of the B-Tree
432 */
433 hammer_volume_t volume;
434
435 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
436 KKASSERT(error == 0);
437
438 hammer_modify_volume_field(cursor->trans, volume,
439 vol0_btree_root);
440 volume->ondisk->vol0_btree_root = nnode->node_offset;
441 hammer_modify_volume_done(volume);
442 hammer_rel_volume(volume, 0);
443 }
444
445 /*
446 * Now adjust our children's pointers to us.
447 */
448 for (i = 0; i < nnode->ondisk->count; ++i) {
449 elm = &nnode->ondisk->elms[i];
450 error = btree_set_parent(cursor->trans, nnode, elm);
451 if (error)
452 panic("reblock internal node: fixup problem");
453 }
454
455 /*
456 * Clean up.
457 *
458 * The new node replaces the current node in the cursor. The cursor
459 * expects it to be locked so leave it locked. Discard onode.
460 */
461 hammer_delete_node(cursor->trans, onode);
462
463 if (hammer_debug_general & 0x4000) {
464 kprintf("REBLOCK INODE %016llx -> %016llx\n",
b58c6388
MD
465 onode->node_offset, nnode->node_offset);
466 }
8d0efe43 467 hammer_modify_node_done(nnode);
bf686dbe 468 cursor->node = nnode;
09ac686b
MD
469
470 hammer_unlock(&onode->lock);
bf686dbe
MD
471 hammer_rel_node(onode);
472
2f85fa4d
MD
473done:
474 hammer_btree_unlock_children(&locklist);
bf686dbe
MD
475 return (error);
476}
477