HAMMER VFS - Fix probable corruption case when filesystem becomes nearly full
[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 *
44a83111 34 * $DragonFly: src/sys/vfs/hammer/hammer_reblock.c,v 1.34 2008/11/13 02:18:43 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;
a7e9bef1 64 int checkspace_count;
93291532
MD
65 int error;
66 int seq;
7b6ccb11
MD
67 int slop;
68
69 /*
70 * A fill level <= 20% is considered an emergency. free_level is
71 * inverted from fill_level.
72 */
73 if (reblock->free_level >= HAMMER_LARGEBLOCK_SIZE * 8 / 10)
74 slop = HAMMER_CHKSPC_EMERGENCY;
75 else
76 slop = HAMMER_CHKSPC_REBLOCK;
bf686dbe 77
dd94f1b1
MD
78 if ((reblock->key_beg.localization | reblock->key_end.localization) &
79 HAMMER_LOCALIZE_PSEUDOFS_MASK) {
80 return(EINVAL);
81 }
82 if (reblock->key_beg.obj_id >= reblock->key_end.obj_id)
bf686dbe
MD
83 return(EINVAL);
84 if (reblock->free_level < 0)
85 return(EINVAL);
86
dd94f1b1 87 reblock->key_cur = reblock->key_beg;
842e7a70 88 reblock->key_cur.localization &= HAMMER_LOCALIZE_MASK;
dd94f1b1 89 reblock->key_cur.localization += ip->obj_localization;
814387f6 90
a7e9bef1 91 checkspace_count = 0;
93291532 92 seq = trans->hmp->flusher.act;
bf686dbe 93retry:
4e17f465 94 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
bf686dbe
MD
95 if (error) {
96 hammer_done_cursor(&cursor);
dd94f1b1 97 goto failed;
bf686dbe 98 }
dd94f1b1
MD
99 cursor.key_beg.localization = reblock->key_cur.localization;
100 cursor.key_beg.obj_id = reblock->key_cur.obj_id;
bf686dbe
MD
101 cursor.key_beg.key = HAMMER_MIN_KEY;
102 cursor.key_beg.create_tid = 1;
103 cursor.key_beg.delete_tid = 0;
104 cursor.key_beg.rec_type = HAMMER_MIN_RECTYPE;
105 cursor.key_beg.obj_type = 0;
106
842e7a70
MD
107 cursor.key_end.localization = (reblock->key_end.localization &
108 HAMMER_LOCALIZE_MASK) +
dd94f1b1
MD
109 ip->obj_localization;
110 cursor.key_end.obj_id = reblock->key_end.obj_id;
bf686dbe
MD
111 cursor.key_end.key = HAMMER_MAX_KEY;
112 cursor.key_end.create_tid = HAMMER_MAX_TID - 1;
113 cursor.key_end.delete_tid = 0;
114 cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
115 cursor.key_end.obj_type = 0;
116
117 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
9480ff55 118 cursor.flags |= HAMMER_CURSOR_BACKEND;
bf686dbe 119
2f85fa4d
MD
120 /*
121 * This flag allows the btree scan code to return internal nodes,
122 * so we can reblock them in addition to the leafs. Only specify it
123 * if we intend to reblock B-Tree nodes.
124 */
125 if (reblock->head.flags & HAMMER_IOC_DO_BTREE)
126 cursor.flags |= HAMMER_CURSOR_REBLOCKING;
127
bf686dbe
MD
128 error = hammer_btree_first(&cursor);
129 while (error == 0) {
2f85fa4d
MD
130 /*
131 * Internal or Leaf node
132 */
07ed04b5 133 KKASSERT(cursor.index < cursor.node->ondisk->count);
bf686dbe 134 elm = &cursor.node->ondisk->elms[cursor.index];
dd94f1b1
MD
135 reblock->key_cur.obj_id = elm->base.obj_id;
136 reblock->key_cur.localization = elm->base.localization;
bf686dbe 137
9480ff55 138 /*
9f5097dc
MD
139 * Yield to more important tasks
140 */
141 if ((error = hammer_signal_check(trans->hmp)) != 0)
142 break;
a7e9bef1
MD
143
144 /*
a7e9bef1
MD
145 * If there is insufficient free space it may be due to
146 * reserved bigblocks, which flushing might fix.
c9ce54d6 147 *
07ed04b5
MD
148 * We must force a retest in case the unlocked cursor is
149 * moved to the end of the leaf, or moved to an internal
150 * node.
151 *
c9ce54d6 152 * WARNING: See warnings in hammer_unlock_cursor() function.
a7e9bef1 153 */
7b6ccb11 154 if (hammer_checkspace(trans->hmp, slop)) {
a7e9bef1
MD
155 if (++checkspace_count == 10) {
156 error = ENOSPC;
93291532 157 break;
a7e9bef1 158 }
982be4bf 159 hammer_unlock_cursor(&cursor);
07ed04b5 160 cursor.flags |= HAMMER_CURSOR_RETEST;
93291532 161 hammer_flusher_wait(trans->hmp, seq);
982be4bf 162 hammer_lock_cursor(&cursor);
7a61b85d 163 seq = hammer_flusher_async(trans->hmp, NULL);
07ed04b5 164 goto skip;
a7e9bef1
MD
165 }
166
167 /*
9480ff55
MD
168 * Acquiring the sync_lock prevents the operation from
169 * crossing a synchronization boundary.
09ac686b
MD
170 *
171 * NOTE: cursor.node may have changed on return.
c9ce54d6
MD
172 *
173 * WARNING: See warnings in hammer_unlock_cursor() function.
9480ff55 174 */
2f85fa4d 175 hammer_sync_lock_sh(trans);
36f82b23 176 error = hammer_reblock_helper(reblock, &cursor, elm);
2f85fa4d 177 hammer_sync_unlock(trans);
93291532 178
15e75dab
MD
179 while (hammer_flusher_meta_halflimit(trans->hmp) ||
180 hammer_flusher_undo_exhausted(trans, 2)) {
982be4bf 181 hammer_unlock_cursor(&cursor);
93291532 182 hammer_flusher_wait(trans->hmp, seq);
982be4bf 183 hammer_lock_cursor(&cursor);
15e75dab 184 seq = hammer_flusher_async_one(trans->hmp);
93291532 185 }
1b0ab2c3
MD
186
187 /*
188 * Setup for iteration, our cursor flags may be modified by
189 * other threads while we are unlocked.
190 */
191 cursor.flags |= HAMMER_CURSOR_ATEDISK;
192
193 /*
194 * We allocate data buffers, which atm we don't track
195 * dirty levels for because we allow the kernel to write
196 * them. But if we allocate too many we can still deadlock
197 * the buffer cache.
198 *
c9ce54d6
MD
199 * WARNING: See warnings in hammer_unlock_cursor() function.
200 * (The cursor's node and element may change!)
1b0ab2c3
MD
201 */
202 if (bd_heatup()) {
982be4bf 203 hammer_unlock_cursor(&cursor);
1b0ab2c3 204 bwillwrite(HAMMER_XBUFSIZE);
982be4bf 205 hammer_lock_cursor(&cursor);
1b0ab2c3 206 }
07ed04b5 207skip:
bf686dbe 208 if (error == 0) {
bf686dbe
MD
209 error = hammer_btree_iterate(&cursor);
210 }
211 }
212 if (error == ENOENT)
213 error = 0;
214 hammer_done_cursor(&cursor);
06ad81ff
MD
215 if (error == EWOULDBLOCK) {
216 hammer_flusher_sync(trans->hmp);
217 goto retry;
218 }
bf686dbe
MD
219 if (error == EDEADLK)
220 goto retry;
19619882
MD
221 if (error == EINTR) {
222 reblock->head.flags |= HAMMER_IOC_HEAD_INTR;
223 error = 0;
224 }
dd94f1b1
MD
225failed:
226 reblock->key_cur.localization &= HAMMER_LOCALIZE_MASK;
bf686dbe
MD
227 return(error);
228}
229
230/*
231 * Reblock the B-Tree (leaf) node, record, and/or data if necessary.
232 *
9480ff55
MD
233 * XXX We have no visibility into internal B-Tree nodes at the moment,
234 * only leaf nodes.
bf686dbe
MD
235 */
236static int
36f82b23 237hammer_reblock_helper(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
238 hammer_cursor_t cursor, hammer_btree_elm_t elm)
239{
43c665ae 240 hammer_mount_t hmp;
bf686dbe 241 hammer_off_t tmp_offset;
ebbcfba9 242 hammer_node_ondisk_t ondisk;
44a83111 243 struct hammer_btree_leaf_elm leaf;
bf686dbe 244 int error;
bf686dbe
MD
245 int bytes;
246 int cur;
bf3b416b 247 int iocflags;
bf686dbe 248
bf686dbe 249 error = 0;
43c665ae 250 hmp = cursor->trans->hmp;
bf686dbe
MD
251
252 /*
253 * Reblock data. Note that data embedded in a record is reblocked
2f85fa4d
MD
254 * by the record reblock code. Data processing only occurs at leaf
255 * nodes and for RECORD element types.
bf686dbe 256 */
2f85fa4d
MD
257 if (cursor->node->ondisk->type != HAMMER_BTREE_TYPE_LEAF)
258 goto skip;
259 if (elm->leaf.base.btype != HAMMER_BTREE_TYPE_RECORD)
260 return(0);
bf686dbe 261 tmp_offset = elm->leaf.data_offset;
bf3b416b
MD
262 if (tmp_offset == 0)
263 goto skip;
264 if (error)
265 goto skip;
266
267 /*
268 * NOTE: Localization restrictions may also have been set-up, we can't
83f2a3aa 269 * just set the match flags willy-nilly here.
bf3b416b
MD
270 */
271 switch(elm->leaf.base.rec_type) {
272 case HAMMER_RECTYPE_INODE:
83f2a3aa
MD
273 case HAMMER_RECTYPE_SNAPSHOT:
274 case HAMMER_RECTYPE_CONFIG:
bf3b416b
MD
275 iocflags = HAMMER_IOC_DO_INODES;
276 break;
277 case HAMMER_RECTYPE_EXT:
278 case HAMMER_RECTYPE_FIX:
ea434b6f 279 case HAMMER_RECTYPE_PFS:
bf3b416b
MD
280 case HAMMER_RECTYPE_DIRENTRY:
281 iocflags = HAMMER_IOC_DO_DIRS;
282 break;
283 case HAMMER_RECTYPE_DATA:
284 case HAMMER_RECTYPE_DB:
285 iocflags = HAMMER_IOC_DO_DATA;
286 break;
287 default:
288 iocflags = 0;
289 break;
290 }
291 if (reblock->head.flags & iocflags) {
bf686dbe
MD
292 ++reblock->data_count;
293 reblock->data_byte_count += elm->leaf.data_len;
43c665ae 294 bytes = hammer_blockmap_getfree(hmp, tmp_offset, &cur, &error);
2f85fa4d
MD
295 if (hammer_debug_general & 0x4000)
296 kprintf("D %6d/%d\n", bytes, reblock->free_level);
bf3b416b
MD
297 if (error == 0 && (cur == 0 || reblock->free_level == 0) &&
298 bytes >= reblock->free_level) {
44a83111
MD
299 /*
300 * This is nasty, the uncache code may have to get
301 * vnode locks and because of that we can't hold
302 * the cursor locked.
c9ce54d6
MD
303 *
304 * WARNING: See warnings in hammer_unlock_cursor()
305 * function.
44a83111
MD
306 */
307 leaf = elm->leaf;
982be4bf 308 hammer_unlock_cursor(cursor);
44a83111 309 hammer_io_direct_uncache(hmp, &leaf);
982be4bf 310 hammer_lock_cursor(cursor);
ebbcfba9
MD
311
312 /*
313 * elm may have become stale or invalid, reload it.
314 * ondisk variable is temporary only. Note that
315 * cursor->node and thus cursor->node->ondisk may
316 * also changed.
317 */
318 ondisk = cursor->node->ondisk;
319 elm = &ondisk->elms[cursor->index];
44a83111 320 if (cursor->flags & HAMMER_CURSOR_RETEST) {
ebbcfba9
MD
321 kprintf("hammer: debug: retest on "
322 "reblocker uncache\n");
323 error = EDEADLK;
324 } else if (ondisk->type != HAMMER_BTREE_TYPE_LEAF ||
325 cursor->index >= ondisk->count) {
326 kprintf("hammer: debug: shifted on "
327 "reblocker uncache\n");
328 error = EDEADLK;
329 } else if (bcmp(&elm->leaf, &leaf, sizeof(leaf))) {
330 kprintf("hammer: debug: changed on "
331 "reblocker uncache\n");
44a83111 332 error = EDEADLK;
44a83111
MD
333 }
334 if (error == 0)
335 error = hammer_cursor_upgrade(cursor);
bf686dbe 336 if (error == 0) {
07ed04b5 337 KKASSERT(cursor->index < ondisk->count);
36f82b23 338 error = hammer_reblock_data(reblock,
bf686dbe
MD
339 cursor, elm);
340 }
341 if (error == 0) {
342 ++reblock->data_moves;
343 reblock->data_byte_moves += elm->leaf.data_len;
344 }
345 }
346 }
347
2f85fa4d 348skip:
bf686dbe 349 /*
1775b6a0
MD
350 * Reblock a B-Tree internal or leaf node. A leaf node is reblocked
351 * on initial entry only (element 0). An internal node is reblocked
352 * when entered upward from its first leaf node only (also element 0).
353 * Further revisits of the internal node (index > 0) are ignored.
bf686dbe
MD
354 */
355 tmp_offset = cursor->node->node_offset;
bf3b416b 356 if (cursor->index == 0 &&
814387f6 357 error == 0 && (reblock->head.flags & HAMMER_IOC_DO_BTREE)) {
bf686dbe 358 ++reblock->btree_count;
43c665ae 359 bytes = hammer_blockmap_getfree(hmp, tmp_offset, &cur, &error);
2f85fa4d
MD
360 if (hammer_debug_general & 0x4000)
361 kprintf("B %6d/%d\n", bytes, reblock->free_level);
bf3b416b
MD
362 if (error == 0 && (cur == 0 || reblock->free_level == 0) &&
363 bytes >= reblock->free_level) {
bf686dbe
MD
364 error = hammer_cursor_upgrade(cursor);
365 if (error == 0) {
07ed04b5
MD
366 if (cursor->parent) {
367 KKASSERT(cursor->parent_index <
368 cursor->parent->ondisk->count);
bf686dbe 369 elm = &cursor->parent->ondisk->elms[cursor->parent_index];
07ed04b5 370 } else {
bf686dbe 371 elm = NULL;
07ed04b5 372 }
2f85fa4d
MD
373 switch(cursor->node->ondisk->type) {
374 case HAMMER_BTREE_TYPE_LEAF:
375 error = hammer_reblock_leaf_node(
376 reblock, cursor, elm);
377 break;
378 case HAMMER_BTREE_TYPE_INTERNAL:
379 error = hammer_reblock_int_node(
380 reblock, cursor, elm);
381 break;
382 default:
383 panic("Illegal B-Tree node type");
384 }
bf686dbe
MD
385 }
386 if (error == 0) {
387 ++reblock->btree_moves;
388 }
389 }
390 }
391
392 hammer_cursor_downgrade(cursor);
393 return(error);
394}
395
396/*
397 * Reblock a record's data. Both the B-Tree element and record pointers
398 * to the data must be adjusted.
399 */
400static int
36f82b23 401hammer_reblock_data(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
402 hammer_cursor_t cursor, hammer_btree_elm_t elm)
403{
404 struct hammer_buffer *data_buffer = NULL;
405 hammer_off_t ndata_offset;
406 int error;
407 void *ndata;
408
409 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA |
11ad5ade 410 HAMMER_CURSOR_GET_LEAF);
bf686dbe
MD
411 if (error)
412 return (error);
36f82b23 413 ndata = hammer_alloc_data(cursor->trans, elm->leaf.data_len,
bf3b416b 414 elm->leaf.base.rec_type,
df2ccbac
MD
415 &ndata_offset, &data_buffer,
416 0, &error);
bf686dbe
MD
417 if (error)
418 goto done;
b8a41159 419 hammer_io_notmeta(data_buffer);
bf686dbe
MD
420
421 /*
422 * Move the data
423 */
10a5d1ba 424 hammer_modify_buffer(cursor->trans, data_buffer, NULL, 0);
bf686dbe 425 bcopy(cursor->data, ndata, elm->leaf.data_len);
10a5d1ba 426 hammer_modify_buffer_done(data_buffer);
bf686dbe 427
36f82b23
MD
428 hammer_blockmap_free(cursor->trans,
429 elm->leaf.data_offset, elm->leaf.data_len);
bf686dbe 430
10a5d1ba
MD
431 hammer_modify_node(cursor->trans, cursor->node,
432 &elm->leaf.data_offset, sizeof(hammer_off_t));
bf686dbe 433 elm->leaf.data_offset = ndata_offset;
10a5d1ba 434 hammer_modify_node_done(cursor->node);
bf686dbe
MD
435
436done:
437 if (data_buffer)
438 hammer_rel_buffer(data_buffer, 0);
439 return (error);
440}
441
442/*
2f85fa4d
MD
443 * Reblock a B-Tree leaf node. The parent must be adjusted to point to
444 * the new copy of the leaf node.
bf686dbe 445 *
2f85fa4d 446 * elm is a pointer to the parent element pointing at cursor.node.
bf686dbe
MD
447 */
448static int
2f85fa4d
MD
449hammer_reblock_leaf_node(struct hammer_ioc_reblock *reblock,
450 hammer_cursor_t cursor, hammer_btree_elm_t elm)
bf686dbe
MD
451{
452 hammer_node_t onode;
453 hammer_node_t nnode;
454 int error;
455
df2ccbac
MD
456 /*
457 * Don't supply a hint when allocating the leaf. Fills are done
458 * from the leaf upwards.
459 */
bf686dbe 460 onode = cursor->node;
df2ccbac 461 nnode = hammer_alloc_btree(cursor->trans, 0, &error);
8d0efe43 462
bf686dbe
MD
463 if (nnode == NULL)
464 return (error);
465
466 /*
467 * Move the node
468 */
09ac686b
MD
469 hammer_lock_ex(&nnode->lock);
470 hammer_modify_node_noundo(cursor->trans, nnode);
bf686dbe
MD
471 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
472
473 if (elm) {
474 /*
475 * We are not the root of the B-Tree
476 */
36f82b23 477 hammer_modify_node(cursor->trans, cursor->parent,
bf686dbe
MD
478 &elm->internal.subtree_offset,
479 sizeof(elm->internal.subtree_offset));
480 elm->internal.subtree_offset = nnode->node_offset;
10a5d1ba 481 hammer_modify_node_done(cursor->parent);
bf686dbe
MD
482 } else {
483 /*
484 * We are the root of the B-Tree
485 */
486 hammer_volume_t volume;
487
36f82b23 488 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
bf686dbe
MD
489 KKASSERT(error == 0);
490
e8599db1
MD
491 hammer_modify_volume_field(cursor->trans, volume,
492 vol0_btree_root);
bf686dbe 493 volume->ondisk->vol0_btree_root = nnode->node_offset;
10a5d1ba 494 hammer_modify_volume_done(volume);
bf686dbe
MD
495 hammer_rel_volume(volume, 0);
496 }
497
b3bad96f 498 hammer_cursor_replaced_node(onode, nnode);
36f82b23 499 hammer_delete_node(cursor->trans, onode);
bf686dbe 500
b58c6388 501 if (hammer_debug_general & 0x4000) {
2f85fa4d 502 kprintf("REBLOCK LNODE %016llx -> %016llx\n",
973c11b9
MD
503 (long long)onode->node_offset,
504 (long long)nnode->node_offset);
2f85fa4d
MD
505 }
506 hammer_modify_node_done(nnode);
507 cursor->node = nnode;
508
509 hammer_unlock(&onode->lock);
510 hammer_rel_node(onode);
511
512 return (error);
513}
514
515/*
516 * Reblock a B-Tree internal node. The parent must be adjusted to point to
517 * the new copy of the internal node, and the node's children's parent
518 * pointers must also be adjusted to point to the new copy.
519 *
520 * elm is a pointer to the parent element pointing at cursor.node.
521 */
522static int
523hammer_reblock_int_node(struct hammer_ioc_reblock *reblock,
524 hammer_cursor_t cursor, hammer_btree_elm_t elm)
525{
1775b6a0 526 struct hammer_node_lock lockroot;
2f85fa4d
MD
527 hammer_node_t onode;
528 hammer_node_t nnode;
df2ccbac 529 hammer_off_t hint;
2f85fa4d
MD
530 int error;
531 int i;
532
1775b6a0 533 hammer_node_lock_init(&lockroot, cursor->node);
24cf83d2 534 error = hammer_btree_lock_children(cursor, 1, &lockroot, NULL);
2f85fa4d
MD
535 if (error)
536 goto done;
537
df2ccbac
MD
538 /*
539 * The internal node is visited after recursing through its
540 * first element. Use the subtree offset allocated for that
541 * element as a hint for allocating the internal node.
542 */
2f85fa4d 543 onode = cursor->node;
df2ccbac
MD
544 if (onode->ondisk->count)
545 hint = onode->ondisk->elms[0].internal.subtree_offset;
546 else
547 hint = 0;
548 nnode = hammer_alloc_btree(cursor->trans, hint, &error);
2f85fa4d
MD
549
550 if (nnode == NULL)
551 goto done;
552
553 /*
554 * Move the node. Adjust the parent's pointer to us first.
555 */
556 hammer_lock_ex(&nnode->lock);
557 hammer_modify_node_noundo(cursor->trans, nnode);
558 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
559
560 if (elm) {
561 /*
562 * We are not the root of the B-Tree
563 */
564 hammer_modify_node(cursor->trans, cursor->parent,
565 &elm->internal.subtree_offset,
566 sizeof(elm->internal.subtree_offset));
567 elm->internal.subtree_offset = nnode->node_offset;
568 hammer_modify_node_done(cursor->parent);
569 } else {
570 /*
571 * We are the root of the B-Tree
572 */
573 hammer_volume_t volume;
574
575 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
576 KKASSERT(error == 0);
577
578 hammer_modify_volume_field(cursor->trans, volume,
579 vol0_btree_root);
580 volume->ondisk->vol0_btree_root = nnode->node_offset;
581 hammer_modify_volume_done(volume);
582 hammer_rel_volume(volume, 0);
583 }
584
585 /*
586 * Now adjust our children's pointers to us.
587 */
588 for (i = 0; i < nnode->ondisk->count; ++i) {
589 elm = &nnode->ondisk->elms[i];
590 error = btree_set_parent(cursor->trans, nnode, elm);
591 if (error)
592 panic("reblock internal node: fixup problem");
593 }
594
595 /*
596 * Clean up.
597 *
598 * The new node replaces the current node in the cursor. The cursor
599 * expects it to be locked so leave it locked. Discard onode.
600 */
b3bad96f 601 hammer_cursor_replaced_node(onode, nnode);
2f85fa4d
MD
602 hammer_delete_node(cursor->trans, onode);
603
604 if (hammer_debug_general & 0x4000) {
605 kprintf("REBLOCK INODE %016llx -> %016llx\n",
973c11b9
MD
606 (long long)onode->node_offset,
607 (long long)nnode->node_offset);
b58c6388 608 }
8d0efe43 609 hammer_modify_node_done(nnode);
bf686dbe 610 cursor->node = nnode;
09ac686b
MD
611
612 hammer_unlock(&onode->lock);
bf686dbe
MD
613 hammer_rel_node(onode);
614
2f85fa4d 615done:
24cf83d2 616 hammer_btree_unlock_children(cursor->trans->hmp, &lockroot, NULL);
bf686dbe
MD
617 return (error);
618}
619