]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
22 | * Copyright 2007 Sun Microsystems, Inc. All rights reserved. | |
23 | * Use is subject to license terms. | |
24 | */ | |
25 | ||
26 | #pragma ident "@(#)dnode_sync.c 1.19 07/08/26 SMI" | |
27 | ||
28 | #include <sys/zfs_context.h> | |
29 | #include <sys/dbuf.h> | |
30 | #include <sys/dnode.h> | |
31 | #include <sys/dmu.h> | |
32 | #include <sys/dmu_tx.h> | |
33 | #include <sys/dmu_objset.h> | |
34 | #include <sys/dsl_dataset.h> | |
35 | #include <sys/spa.h> | |
36 | ||
37 | static void | |
38 | dnode_increase_indirection(dnode_t *dn, dmu_tx_t *tx) | |
39 | { | |
40 | dmu_buf_impl_t *db; | |
41 | int txgoff = tx->tx_txg & TXG_MASK; | |
42 | int nblkptr = dn->dn_phys->dn_nblkptr; | |
43 | int old_toplvl = dn->dn_phys->dn_nlevels - 1; | |
44 | int new_level = dn->dn_next_nlevels[txgoff]; | |
45 | int i; | |
46 | ||
47 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
48 | ||
49 | /* this dnode can't be paged out because it's dirty */ | |
50 | ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); | |
51 | ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); | |
52 | ASSERT(new_level > 1 && dn->dn_phys->dn_nlevels > 0); | |
53 | ||
54 | db = dbuf_hold_level(dn, dn->dn_phys->dn_nlevels, 0, FTAG); | |
55 | ASSERT(db != NULL); | |
56 | ||
57 | dn->dn_phys->dn_nlevels = new_level; | |
58 | dprintf("os=%p obj=%llu, increase to %d\n", dn->dn_objset, | |
59 | dn->dn_object, dn->dn_phys->dn_nlevels); | |
60 | ||
61 | /* check for existing blkptrs in the dnode */ | |
62 | for (i = 0; i < nblkptr; i++) | |
63 | if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[i])) | |
64 | break; | |
65 | if (i != nblkptr) { | |
66 | /* transfer dnode's block pointers to new indirect block */ | |
67 | (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); | |
68 | ASSERT(db->db.db_data); | |
69 | ASSERT(arc_released(db->db_buf)); | |
70 | ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); | |
71 | bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, | |
72 | sizeof (blkptr_t) * nblkptr); | |
73 | arc_buf_freeze(db->db_buf); | |
74 | } | |
75 | ||
76 | /* set dbuf's parent pointers to new indirect buf */ | |
77 | for (i = 0; i < nblkptr; i++) { | |
78 | dmu_buf_impl_t *child = dbuf_find(dn, old_toplvl, i); | |
79 | ||
80 | if (child == NULL) | |
81 | continue; | |
82 | ASSERT3P(child->db_dnode, ==, dn); | |
83 | if (child->db_parent && child->db_parent != dn->dn_dbuf) { | |
84 | ASSERT(child->db_parent->db_level == db->db_level); | |
85 | ASSERT(child->db_blkptr != | |
86 | &dn->dn_phys->dn_blkptr[child->db_blkid]); | |
87 | mutex_exit(&child->db_mtx); | |
88 | continue; | |
89 | } | |
90 | ASSERT(child->db_parent == NULL || | |
91 | child->db_parent == dn->dn_dbuf); | |
92 | ||
93 | child->db_parent = db; | |
94 | dbuf_add_ref(db, child); | |
95 | if (db->db.db_data) | |
96 | child->db_blkptr = (blkptr_t *)db->db.db_data + i; | |
97 | else | |
98 | child->db_blkptr = NULL; | |
99 | dprintf_dbuf_bp(child, child->db_blkptr, | |
100 | "changed db_blkptr to new indirect %s", ""); | |
101 | ||
102 | mutex_exit(&child->db_mtx); | |
103 | } | |
104 | ||
105 | bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); | |
106 | ||
107 | dbuf_rele(db, FTAG); | |
108 | ||
109 | rw_exit(&dn->dn_struct_rwlock); | |
110 | } | |
111 | ||
112 | static void | |
113 | free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) | |
114 | { | |
115 | objset_impl_t *os = dn->dn_objset; | |
116 | uint64_t bytesfreed = 0; | |
117 | int i; | |
118 | ||
119 | dprintf("os=%p obj=%llx num=%d\n", os, dn->dn_object, num); | |
120 | ||
121 | for (i = 0; i < num; i++, bp++) { | |
122 | if (BP_IS_HOLE(bp)) | |
123 | continue; | |
124 | ||
125 | bytesfreed += bp_get_dasize(os->os_spa, bp); | |
126 | ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); | |
127 | dsl_dataset_block_kill(os->os_dsl_dataset, bp, dn->dn_zio, tx); | |
128 | bzero(bp, sizeof (blkptr_t)); | |
129 | } | |
130 | dnode_diduse_space(dn, -bytesfreed); | |
131 | } | |
132 | ||
133 | #ifdef ZFS_DEBUG | |
134 | static void | |
135 | free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) | |
136 | { | |
137 | int off, num; | |
138 | int i, err, epbs; | |
139 | uint64_t txg = tx->tx_txg; | |
140 | ||
141 | epbs = db->db_dnode->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
142 | off = start - (db->db_blkid * 1<<epbs); | |
143 | num = end - start + 1; | |
144 | ||
145 | ASSERT3U(off, >=, 0); | |
146 | ASSERT3U(num, >=, 0); | |
147 | ASSERT3U(db->db_level, >, 0); | |
148 | ASSERT3U(db->db.db_size, ==, 1<<db->db_dnode->dn_phys->dn_indblkshift); | |
149 | ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); | |
150 | ASSERT(db->db_blkptr != NULL); | |
151 | ||
152 | for (i = off; i < off+num; i++) { | |
153 | uint64_t *buf; | |
154 | dmu_buf_impl_t *child; | |
155 | dbuf_dirty_record_t *dr; | |
156 | int j; | |
157 | ||
158 | ASSERT(db->db_level == 1); | |
159 | ||
160 | rw_enter(&db->db_dnode->dn_struct_rwlock, RW_READER); | |
161 | err = dbuf_hold_impl(db->db_dnode, db->db_level-1, | |
162 | (db->db_blkid << epbs) + i, TRUE, FTAG, &child); | |
163 | rw_exit(&db->db_dnode->dn_struct_rwlock); | |
164 | if (err == ENOENT) | |
165 | continue; | |
166 | ASSERT(err == 0); | |
167 | ASSERT(child->db_level == 0); | |
168 | dr = child->db_last_dirty; | |
169 | while (dr && dr->dr_txg > txg) | |
170 | dr = dr->dr_next; | |
171 | ASSERT(dr == NULL || dr->dr_txg == txg); | |
172 | ||
173 | /* data_old better be zeroed */ | |
174 | if (dr) { | |
175 | buf = dr->dt.dl.dr_data->b_data; | |
176 | for (j = 0; j < child->db.db_size >> 3; j++) { | |
177 | if (buf[j] != 0) { | |
178 | panic("freed data not zero: " | |
179 | "child=%p i=%d off=%d num=%d\n", | |
180 | child, i, off, num); | |
181 | } | |
182 | } | |
183 | } | |
184 | ||
185 | /* | |
186 | * db_data better be zeroed unless it's dirty in a | |
187 | * future txg. | |
188 | */ | |
189 | mutex_enter(&child->db_mtx); | |
190 | buf = child->db.db_data; | |
191 | if (buf != NULL && child->db_state != DB_FILL && | |
192 | child->db_last_dirty == NULL) { | |
193 | for (j = 0; j < child->db.db_size >> 3; j++) { | |
194 | if (buf[j] != 0) { | |
195 | panic("freed data not zero: " | |
196 | "child=%p i=%d off=%d num=%d\n", | |
197 | child, i, off, num); | |
198 | } | |
199 | } | |
200 | } | |
201 | mutex_exit(&child->db_mtx); | |
202 | ||
203 | dbuf_rele(child, FTAG); | |
204 | } | |
205 | } | |
206 | #endif | |
207 | ||
208 | static int | |
209 | free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, int trunc, | |
210 | dmu_tx_t *tx) | |
211 | { | |
212 | dnode_t *dn = db->db_dnode; | |
213 | blkptr_t *bp; | |
214 | dmu_buf_impl_t *subdb; | |
215 | uint64_t start, end, dbstart, dbend, i; | |
216 | int epbs, shift, err; | |
217 | int all = TRUE; | |
218 | ||
219 | (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); | |
220 | arc_release(db->db_buf, db); | |
221 | bp = (blkptr_t *)db->db.db_data; | |
222 | ||
223 | epbs = db->db_dnode->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
224 | shift = (db->db_level - 1) * epbs; | |
225 | dbstart = db->db_blkid << epbs; | |
226 | start = blkid >> shift; | |
227 | if (dbstart < start) { | |
228 | bp += start - dbstart; | |
229 | all = FALSE; | |
230 | } else { | |
231 | start = dbstart; | |
232 | } | |
233 | dbend = ((db->db_blkid + 1) << epbs) - 1; | |
234 | end = (blkid + nblks - 1) >> shift; | |
235 | if (dbend <= end) | |
236 | end = dbend; | |
237 | else if (all) | |
238 | all = trunc; | |
239 | ASSERT3U(start, <=, end); | |
240 | ||
241 | if (db->db_level == 1) { | |
242 | FREE_VERIFY(db, start, end, tx); | |
243 | free_blocks(dn, bp, end-start+1, tx); | |
244 | arc_buf_freeze(db->db_buf); | |
245 | ASSERT(all || db->db_last_dirty); | |
246 | return (all); | |
247 | } | |
248 | ||
249 | for (i = start; i <= end; i++, bp++) { | |
250 | if (BP_IS_HOLE(bp)) | |
251 | continue; | |
252 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
253 | err = dbuf_hold_impl(dn, db->db_level-1, i, TRUE, FTAG, &subdb); | |
254 | ASSERT3U(err, ==, 0); | |
255 | rw_exit(&dn->dn_struct_rwlock); | |
256 | ||
257 | if (free_children(subdb, blkid, nblks, trunc, tx)) { | |
258 | ASSERT3P(subdb->db_blkptr, ==, bp); | |
259 | free_blocks(dn, bp, 1, tx); | |
260 | } else { | |
261 | all = FALSE; | |
262 | } | |
263 | dbuf_rele(subdb, FTAG); | |
264 | } | |
265 | arc_buf_freeze(db->db_buf); | |
266 | #ifdef ZFS_DEBUG | |
267 | bp -= (end-start)+1; | |
268 | for (i = start; i <= end; i++, bp++) { | |
269 | if (i == start && blkid != 0) | |
270 | continue; | |
271 | else if (i == end && !trunc) | |
272 | continue; | |
273 | ASSERT3U(bp->blk_birth, ==, 0); | |
274 | } | |
275 | #endif | |
276 | ASSERT(all || db->db_last_dirty); | |
277 | return (all); | |
278 | } | |
279 | ||
280 | /* | |
281 | * free_range: Traverse the indicated range of the provided file | |
282 | * and "free" all the blocks contained there. | |
283 | */ | |
284 | static void | |
285 | dnode_sync_free_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) | |
286 | { | |
287 | blkptr_t *bp = dn->dn_phys->dn_blkptr; | |
288 | dmu_buf_impl_t *db; | |
289 | int trunc, start, end, shift, i, err; | |
290 | int dnlevel = dn->dn_phys->dn_nlevels; | |
291 | ||
292 | if (blkid > dn->dn_phys->dn_maxblkid) | |
293 | return; | |
294 | ||
295 | ASSERT(dn->dn_phys->dn_maxblkid < UINT64_MAX); | |
296 | trunc = blkid + nblks > dn->dn_phys->dn_maxblkid; | |
297 | if (trunc) | |
298 | nblks = dn->dn_phys->dn_maxblkid - blkid + 1; | |
299 | ||
300 | /* There are no indirect blocks in the object */ | |
301 | if (dnlevel == 1) { | |
302 | if (blkid >= dn->dn_phys->dn_nblkptr) { | |
303 | /* this range was never made persistent */ | |
304 | return; | |
305 | } | |
306 | ASSERT3U(blkid + nblks, <=, dn->dn_phys->dn_nblkptr); | |
307 | free_blocks(dn, bp + blkid, nblks, tx); | |
308 | if (trunc) { | |
309 | uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * | |
310 | (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
311 | dn->dn_phys->dn_maxblkid = (blkid ? blkid - 1 : 0); | |
312 | ASSERT(off < dn->dn_phys->dn_maxblkid || | |
313 | dn->dn_phys->dn_maxblkid == 0 || | |
314 | dnode_next_offset(dn, FALSE, &off, | |
315 | 1, 1, 0) != 0); | |
316 | } | |
317 | return; | |
318 | } | |
319 | ||
320 | shift = (dnlevel - 1) * (dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT); | |
321 | start = blkid >> shift; | |
322 | ASSERT(start < dn->dn_phys->dn_nblkptr); | |
323 | end = (blkid + nblks - 1) >> shift; | |
324 | bp += start; | |
325 | for (i = start; i <= end; i++, bp++) { | |
326 | if (BP_IS_HOLE(bp)) | |
327 | continue; | |
328 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
329 | err = dbuf_hold_impl(dn, dnlevel-1, i, TRUE, FTAG, &db); | |
330 | ASSERT3U(err, ==, 0); | |
331 | rw_exit(&dn->dn_struct_rwlock); | |
332 | ||
333 | if (free_children(db, blkid, nblks, trunc, tx)) { | |
334 | ASSERT3P(db->db_blkptr, ==, bp); | |
335 | free_blocks(dn, bp, 1, tx); | |
336 | } | |
337 | dbuf_rele(db, FTAG); | |
338 | } | |
339 | if (trunc) { | |
340 | uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * | |
341 | (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
342 | dn->dn_phys->dn_maxblkid = (blkid ? blkid - 1 : 0); | |
343 | ASSERT(off < dn->dn_phys->dn_maxblkid || | |
344 | dn->dn_phys->dn_maxblkid == 0 || | |
345 | dnode_next_offset(dn, FALSE, &off, 1, 1, 0) != 0); | |
346 | } | |
347 | } | |
348 | ||
349 | /* | |
350 | * Try to kick all the dnodes dbufs out of the cache... | |
351 | */ | |
352 | void | |
353 | dnode_evict_dbufs(dnode_t *dn) | |
354 | { | |
355 | int progress; | |
356 | int pass = 0; | |
357 | ||
358 | do { | |
359 | dmu_buf_impl_t *db, marker; | |
360 | int evicting = FALSE; | |
361 | ||
362 | progress = FALSE; | |
363 | mutex_enter(&dn->dn_dbufs_mtx); | |
364 | list_insert_tail(&dn->dn_dbufs, &marker); | |
365 | db = list_head(&dn->dn_dbufs); | |
366 | for (; db != ▮ db = list_head(&dn->dn_dbufs)) { | |
367 | list_remove(&dn->dn_dbufs, db); | |
368 | list_insert_tail(&dn->dn_dbufs, db); | |
369 | ASSERT3P(db->db_dnode, ==, dn); | |
370 | ||
371 | mutex_enter(&db->db_mtx); | |
372 | if (db->db_state == DB_EVICTING) { | |
373 | progress = TRUE; | |
374 | evicting = TRUE; | |
375 | mutex_exit(&db->db_mtx); | |
376 | } else if (refcount_is_zero(&db->db_holds)) { | |
377 | progress = TRUE; | |
378 | ASSERT(!arc_released(db->db_buf)); | |
379 | dbuf_clear(db); /* exits db_mtx for us */ | |
380 | } else { | |
381 | mutex_exit(&db->db_mtx); | |
382 | } | |
383 | ||
384 | } | |
385 | list_remove(&dn->dn_dbufs, &marker); | |
386 | /* | |
387 | * NB: we need to drop dn_dbufs_mtx between passes so | |
388 | * that any DB_EVICTING dbufs can make progress. | |
389 | * Ideally, we would have some cv we could wait on, but | |
390 | * since we don't, just wait a bit to give the other | |
391 | * thread a chance to run. | |
392 | */ | |
393 | mutex_exit(&dn->dn_dbufs_mtx); | |
394 | if (evicting) | |
395 | delay(1); | |
396 | pass++; | |
397 | ASSERT(pass < 100); /* sanity check */ | |
398 | } while (progress); | |
399 | ||
400 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
401 | if (dn->dn_bonus && refcount_is_zero(&dn->dn_bonus->db_holds)) { | |
402 | mutex_enter(&dn->dn_bonus->db_mtx); | |
403 | dbuf_evict(dn->dn_bonus); | |
404 | dn->dn_bonus = NULL; | |
405 | } | |
406 | rw_exit(&dn->dn_struct_rwlock); | |
407 | } | |
408 | ||
409 | static void | |
410 | dnode_undirty_dbufs(list_t *list) | |
411 | { | |
412 | dbuf_dirty_record_t *dr; | |
413 | ||
414 | while (dr = list_head(list)) { | |
415 | dmu_buf_impl_t *db = dr->dr_dbuf; | |
416 | uint64_t txg = dr->dr_txg; | |
417 | ||
418 | mutex_enter(&db->db_mtx); | |
419 | /* XXX - use dbuf_undirty()? */ | |
420 | list_remove(list, dr); | |
421 | ASSERT(db->db_last_dirty == dr); | |
422 | db->db_last_dirty = NULL; | |
423 | db->db_dirtycnt -= 1; | |
424 | if (db->db_level == 0) { | |
425 | ASSERT(db->db_blkid == DB_BONUS_BLKID || | |
426 | dr->dt.dl.dr_data == db->db_buf); | |
427 | dbuf_unoverride(dr); | |
428 | mutex_exit(&db->db_mtx); | |
429 | } else { | |
430 | mutex_exit(&db->db_mtx); | |
431 | dnode_undirty_dbufs(&dr->dt.di.dr_children); | |
432 | } | |
433 | kmem_free(dr, sizeof (dbuf_dirty_record_t)); | |
434 | dbuf_rele(db, (void *)(uintptr_t)txg); | |
435 | } | |
436 | } | |
437 | ||
438 | static void | |
439 | dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) | |
440 | { | |
441 | int txgoff = tx->tx_txg & TXG_MASK; | |
442 | ||
443 | ASSERT(dmu_tx_is_syncing(tx)); | |
444 | ||
445 | dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); | |
446 | dnode_evict_dbufs(dn); | |
447 | ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); | |
448 | ||
449 | /* | |
450 | * XXX - It would be nice to assert this, but we may still | |
451 | * have residual holds from async evictions from the arc... | |
452 | * | |
453 | * zfs_obj_to_path() also depends on this being | |
454 | * commented out. | |
455 | * | |
456 | * ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); | |
457 | */ | |
458 | ||
459 | /* Undirty next bits */ | |
460 | dn->dn_next_nlevels[txgoff] = 0; | |
461 | dn->dn_next_indblkshift[txgoff] = 0; | |
462 | dn->dn_next_blksz[txgoff] = 0; | |
463 | ||
464 | /* free up all the blocks in the file. */ | |
465 | dnode_sync_free_range(dn, 0, dn->dn_phys->dn_maxblkid+1, tx); | |
466 | ASSERT3U(DN_USED_BYTES(dn->dn_phys), ==, 0); | |
467 | ||
468 | /* ASSERT(blkptrs are zero); */ | |
469 | ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); | |
470 | ASSERT(dn->dn_type != DMU_OT_NONE); | |
471 | ||
472 | ASSERT(dn->dn_free_txg > 0); | |
473 | if (dn->dn_allocated_txg != dn->dn_free_txg) | |
474 | dbuf_will_dirty(dn->dn_dbuf, tx); | |
475 | bzero(dn->dn_phys, sizeof (dnode_phys_t)); | |
476 | ||
477 | mutex_enter(&dn->dn_mtx); | |
478 | dn->dn_type = DMU_OT_NONE; | |
479 | dn->dn_maxblkid = 0; | |
480 | dn->dn_allocated_txg = 0; | |
481 | dn->dn_free_txg = 0; | |
482 | mutex_exit(&dn->dn_mtx); | |
483 | ||
484 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); | |
485 | ||
486 | dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); | |
487 | /* | |
488 | * Now that we've released our hold, the dnode may | |
489 | * be evicted, so we musn't access it. | |
490 | */ | |
491 | } | |
492 | ||
493 | /* | |
494 | * Write out the dnode's dirty buffers. | |
495 | * | |
496 | * NOTE: The dnode is kept in memory by being dirty. Once the | |
497 | * dirty bit is cleared, it may be evicted. Beware of this! | |
498 | */ | |
499 | void | |
500 | dnode_sync(dnode_t *dn, dmu_tx_t *tx) | |
501 | { | |
502 | free_range_t *rp; | |
503 | dnode_phys_t *dnp = dn->dn_phys; | |
504 | int txgoff = tx->tx_txg & TXG_MASK; | |
505 | list_t *list = &dn->dn_dirty_records[txgoff]; | |
506 | ||
507 | ASSERT(dmu_tx_is_syncing(tx)); | |
508 | ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg); | |
509 | DNODE_VERIFY(dn); | |
510 | ||
511 | ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf)); | |
512 | ||
513 | mutex_enter(&dn->dn_mtx); | |
514 | if (dn->dn_allocated_txg == tx->tx_txg) { | |
515 | /* The dnode is newly allocated or reallocated */ | |
516 | if (dnp->dn_type == DMU_OT_NONE) { | |
517 | /* this is a first alloc, not a realloc */ | |
518 | /* XXX shouldn't the phys already be zeroed? */ | |
519 | bzero(dnp, DNODE_CORE_SIZE); | |
520 | dnp->dn_nlevels = 1; | |
521 | } | |
522 | ||
523 | if (dn->dn_nblkptr > dnp->dn_nblkptr) { | |
524 | /* zero the new blkptrs we are gaining */ | |
525 | bzero(dnp->dn_blkptr + dnp->dn_nblkptr, | |
526 | sizeof (blkptr_t) * | |
527 | (dn->dn_nblkptr - dnp->dn_nblkptr)); | |
528 | } | |
529 | dnp->dn_type = dn->dn_type; | |
530 | dnp->dn_bonustype = dn->dn_bonustype; | |
531 | dnp->dn_bonuslen = dn->dn_bonuslen; | |
532 | dnp->dn_nblkptr = dn->dn_nblkptr; | |
533 | } | |
534 | ||
535 | ASSERT(dnp->dn_nlevels > 1 || | |
536 | BP_IS_HOLE(&dnp->dn_blkptr[0]) || | |
537 | BP_GET_LSIZE(&dnp->dn_blkptr[0]) == | |
538 | dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
539 | ||
540 | if (dn->dn_next_blksz[txgoff]) { | |
541 | ASSERT(P2PHASE(dn->dn_next_blksz[txgoff], | |
542 | SPA_MINBLOCKSIZE) == 0); | |
543 | ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[0]) || | |
544 | list_head(list) != NULL || | |
545 | dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT == | |
546 | dnp->dn_datablkszsec); | |
547 | dnp->dn_datablkszsec = | |
548 | dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT; | |
549 | dn->dn_next_blksz[txgoff] = 0; | |
550 | } | |
551 | ||
552 | if (dn->dn_next_bonuslen[txgoff]) { | |
553 | if (dn->dn_next_bonuslen[txgoff] == DN_ZERO_BONUSLEN) | |
554 | dnp->dn_bonuslen = 0; | |
555 | else | |
556 | dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff]; | |
557 | ASSERT(dnp->dn_bonuslen <= DN_MAX_BONUSLEN); | |
558 | dn->dn_next_bonuslen[txgoff] = 0; | |
559 | } | |
560 | ||
561 | if (dn->dn_next_indblkshift[txgoff]) { | |
562 | ASSERT(dnp->dn_nlevels == 1); | |
563 | dnp->dn_indblkshift = dn->dn_next_indblkshift[txgoff]; | |
564 | dn->dn_next_indblkshift[txgoff] = 0; | |
565 | } | |
566 | ||
567 | /* | |
568 | * Just take the live (open-context) values for checksum and compress. | |
569 | * Strictly speaking it's a future leak, but nothing bad happens if we | |
570 | * start using the new checksum or compress algorithm a little early. | |
571 | */ | |
572 | dnp->dn_checksum = dn->dn_checksum; | |
573 | dnp->dn_compress = dn->dn_compress; | |
574 | ||
575 | mutex_exit(&dn->dn_mtx); | |
576 | ||
577 | /* process all the "freed" ranges in the file */ | |
578 | if (dn->dn_free_txg == 0 || dn->dn_free_txg > tx->tx_txg) { | |
579 | for (rp = avl_last(&dn->dn_ranges[txgoff]); rp != NULL; | |
580 | rp = AVL_PREV(&dn->dn_ranges[txgoff], rp)) | |
581 | dnode_sync_free_range(dn, | |
582 | rp->fr_blkid, rp->fr_nblks, tx); | |
583 | } | |
584 | /* grab the mutex so we don't race with dnode_block_freed() */ | |
585 | mutex_enter(&dn->dn_mtx); | |
586 | for (rp = avl_first(&dn->dn_ranges[txgoff]); rp; ) { | |
587 | ||
588 | free_range_t *last = rp; | |
589 | rp = AVL_NEXT(&dn->dn_ranges[txgoff], rp); | |
590 | avl_remove(&dn->dn_ranges[txgoff], last); | |
591 | kmem_free(last, sizeof (free_range_t)); | |
592 | } | |
593 | mutex_exit(&dn->dn_mtx); | |
594 | if (dn->dn_free_txg > 0 && dn->dn_free_txg <= tx->tx_txg) { | |
595 | dnode_sync_free(dn, tx); | |
596 | return; | |
597 | } | |
598 | ||
599 | if (dn->dn_next_nlevels[txgoff]) { | |
600 | dnode_increase_indirection(dn, tx); | |
601 | dn->dn_next_nlevels[txgoff] = 0; | |
602 | } | |
603 | ||
604 | dbuf_sync_list(list, tx); | |
605 | ||
606 | if (dn->dn_object != DMU_META_DNODE_OBJECT) { | |
607 | ASSERT3P(list_head(list), ==, NULL); | |
608 | dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); | |
609 | } | |
610 | ||
611 | /* | |
612 | * Although we have dropped our reference to the dnode, it | |
613 | * can't be evicted until its written, and we haven't yet | |
614 | * initiated the IO for the dnode's dbuf. | |
615 | */ | |
616 | } |