]>
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 | /* | |
d164b209 | 22 | * Copyright 2009 Sun Microsystems, Inc. All rights reserved. |
34dc7c2f BB |
23 | * Use is subject to license terms. |
24 | */ | |
25 | ||
34dc7c2f BB |
26 | #include <sys/zfs_context.h> |
27 | #include <sys/dbuf.h> | |
28 | #include <sys/dnode.h> | |
29 | #include <sys/dmu.h> | |
30 | #include <sys/dmu_tx.h> | |
31 | #include <sys/dmu_objset.h> | |
32 | #include <sys/dsl_dataset.h> | |
33 | #include <sys/spa.h> | |
34 | ||
35 | static void | |
36 | dnode_increase_indirection(dnode_t *dn, dmu_tx_t *tx) | |
37 | { | |
38 | dmu_buf_impl_t *db; | |
39 | int txgoff = tx->tx_txg & TXG_MASK; | |
40 | int nblkptr = dn->dn_phys->dn_nblkptr; | |
41 | int old_toplvl = dn->dn_phys->dn_nlevels - 1; | |
42 | int new_level = dn->dn_next_nlevels[txgoff]; | |
43 | int i; | |
44 | ||
45 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
46 | ||
47 | /* this dnode can't be paged out because it's dirty */ | |
48 | ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); | |
49 | ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); | |
50 | ASSERT(new_level > 1 && dn->dn_phys->dn_nlevels > 0); | |
51 | ||
52 | db = dbuf_hold_level(dn, dn->dn_phys->dn_nlevels, 0, FTAG); | |
53 | ASSERT(db != NULL); | |
54 | ||
55 | dn->dn_phys->dn_nlevels = new_level; | |
56 | dprintf("os=%p obj=%llu, increase to %d\n", dn->dn_objset, | |
57 | dn->dn_object, dn->dn_phys->dn_nlevels); | |
58 | ||
59 | /* check for existing blkptrs in the dnode */ | |
60 | for (i = 0; i < nblkptr; i++) | |
61 | if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[i])) | |
62 | break; | |
63 | if (i != nblkptr) { | |
64 | /* transfer dnode's block pointers to new indirect block */ | |
65 | (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); | |
66 | ASSERT(db->db.db_data); | |
67 | ASSERT(arc_released(db->db_buf)); | |
68 | ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); | |
69 | bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, | |
70 | sizeof (blkptr_t) * nblkptr); | |
71 | arc_buf_freeze(db->db_buf); | |
72 | } | |
73 | ||
74 | /* set dbuf's parent pointers to new indirect buf */ | |
75 | for (i = 0; i < nblkptr; i++) { | |
76 | dmu_buf_impl_t *child = dbuf_find(dn, old_toplvl, i); | |
77 | ||
78 | if (child == NULL) | |
79 | continue; | |
80 | ASSERT3P(child->db_dnode, ==, dn); | |
81 | if (child->db_parent && child->db_parent != dn->dn_dbuf) { | |
82 | ASSERT(child->db_parent->db_level == db->db_level); | |
83 | ASSERT(child->db_blkptr != | |
84 | &dn->dn_phys->dn_blkptr[child->db_blkid]); | |
85 | mutex_exit(&child->db_mtx); | |
86 | continue; | |
87 | } | |
88 | ASSERT(child->db_parent == NULL || | |
89 | child->db_parent == dn->dn_dbuf); | |
90 | ||
91 | child->db_parent = db; | |
92 | dbuf_add_ref(db, child); | |
93 | if (db->db.db_data) | |
94 | child->db_blkptr = (blkptr_t *)db->db.db_data + i; | |
95 | else | |
96 | child->db_blkptr = NULL; | |
97 | dprintf_dbuf_bp(child, child->db_blkptr, | |
98 | "changed db_blkptr to new indirect %s", ""); | |
99 | ||
100 | mutex_exit(&child->db_mtx); | |
101 | } | |
102 | ||
103 | bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); | |
104 | ||
105 | dbuf_rele(db, FTAG); | |
106 | ||
107 | rw_exit(&dn->dn_struct_rwlock); | |
108 | } | |
109 | ||
b128c09f | 110 | static int |
34dc7c2f BB |
111 | free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) |
112 | { | |
b128c09f | 113 | dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; |
34dc7c2f | 114 | uint64_t bytesfreed = 0; |
b128c09f | 115 | int i, blocks_freed = 0; |
34dc7c2f | 116 | |
b128c09f | 117 | dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num); |
34dc7c2f BB |
118 | |
119 | for (i = 0; i < num; i++, bp++) { | |
120 | if (BP_IS_HOLE(bp)) | |
121 | continue; | |
122 | ||
b128c09f | 123 | bytesfreed += dsl_dataset_block_kill(ds, bp, dn->dn_zio, tx); |
34dc7c2f | 124 | ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); |
34dc7c2f | 125 | bzero(bp, sizeof (blkptr_t)); |
b128c09f | 126 | blocks_freed += 1; |
34dc7c2f BB |
127 | } |
128 | dnode_diduse_space(dn, -bytesfreed); | |
b128c09f | 129 | return (blocks_freed); |
34dc7c2f BB |
130 | } |
131 | ||
132 | #ifdef ZFS_DEBUG | |
133 | static void | |
134 | free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) | |
135 | { | |
136 | int off, num; | |
137 | int i, err, epbs; | |
138 | uint64_t txg = tx->tx_txg; | |
139 | ||
140 | epbs = db->db_dnode->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
141 | off = start - (db->db_blkid * 1<<epbs); | |
142 | num = end - start + 1; | |
143 | ||
144 | ASSERT3U(off, >=, 0); | |
145 | ASSERT3U(num, >=, 0); | |
146 | ASSERT3U(db->db_level, >, 0); | |
147 | ASSERT3U(db->db.db_size, ==, 1<<db->db_dnode->dn_phys->dn_indblkshift); | |
148 | ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); | |
149 | ASSERT(db->db_blkptr != NULL); | |
150 | ||
151 | for (i = off; i < off+num; i++) { | |
152 | uint64_t *buf; | |
153 | dmu_buf_impl_t *child; | |
154 | dbuf_dirty_record_t *dr; | |
155 | int j; | |
156 | ||
157 | ASSERT(db->db_level == 1); | |
158 | ||
159 | rw_enter(&db->db_dnode->dn_struct_rwlock, RW_READER); | |
160 | err = dbuf_hold_impl(db->db_dnode, db->db_level-1, | |
161 | (db->db_blkid << epbs) + i, TRUE, FTAG, &child); | |
162 | rw_exit(&db->db_dnode->dn_struct_rwlock); | |
163 | if (err == ENOENT) | |
164 | continue; | |
165 | ASSERT(err == 0); | |
166 | ASSERT(child->db_level == 0); | |
167 | dr = child->db_last_dirty; | |
168 | while (dr && dr->dr_txg > txg) | |
169 | dr = dr->dr_next; | |
170 | ASSERT(dr == NULL || dr->dr_txg == txg); | |
171 | ||
172 | /* data_old better be zeroed */ | |
173 | if (dr) { | |
174 | buf = dr->dt.dl.dr_data->b_data; | |
175 | for (j = 0; j < child->db.db_size >> 3; j++) { | |
176 | if (buf[j] != 0) { | |
177 | panic("freed data not zero: " | |
178 | "child=%p i=%d off=%d num=%d\n", | |
b128c09f | 179 | (void *)child, i, off, num); |
34dc7c2f BB |
180 | } |
181 | } | |
182 | } | |
183 | ||
184 | /* | |
185 | * db_data better be zeroed unless it's dirty in a | |
186 | * future txg. | |
187 | */ | |
188 | mutex_enter(&child->db_mtx); | |
189 | buf = child->db.db_data; | |
190 | if (buf != NULL && child->db_state != DB_FILL && | |
191 | child->db_last_dirty == NULL) { | |
192 | for (j = 0; j < child->db.db_size >> 3; j++) { | |
193 | if (buf[j] != 0) { | |
194 | panic("freed data not zero: " | |
195 | "child=%p i=%d off=%d num=%d\n", | |
b128c09f | 196 | (void *)child, i, off, num); |
34dc7c2f BB |
197 | } |
198 | } | |
199 | } | |
200 | mutex_exit(&child->db_mtx); | |
201 | ||
202 | dbuf_rele(child, FTAG); | |
203 | } | |
204 | } | |
205 | #endif | |
206 | ||
b128c09f BB |
207 | #define ALL -1 |
208 | ||
34dc7c2f BB |
209 | static int |
210 | free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, int trunc, | |
211 | dmu_tx_t *tx) | |
212 | { | |
213 | dnode_t *dn = db->db_dnode; | |
214 | blkptr_t *bp; | |
215 | dmu_buf_impl_t *subdb; | |
216 | uint64_t start, end, dbstart, dbend, i; | |
217 | int epbs, shift, err; | |
218 | int all = TRUE; | |
b128c09f BB |
219 | int blocks_freed = 0; |
220 | ||
221 | /* | |
222 | * There is a small possibility that this block will not be cached: | |
223 | * 1 - if level > 1 and there are no children with level <= 1 | |
224 | * 2 - if we didn't get a dirty hold (because this block had just | |
225 | * finished being written -- and so had no holds), and then this | |
226 | * block got evicted before we got here. | |
227 | */ | |
228 | if (db->db_state != DB_CACHED) | |
229 | (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); | |
34dc7c2f | 230 | |
34dc7c2f BB |
231 | arc_release(db->db_buf, db); |
232 | bp = (blkptr_t *)db->db.db_data; | |
233 | ||
234 | epbs = db->db_dnode->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
235 | shift = (db->db_level - 1) * epbs; | |
236 | dbstart = db->db_blkid << epbs; | |
237 | start = blkid >> shift; | |
238 | if (dbstart < start) { | |
239 | bp += start - dbstart; | |
240 | all = FALSE; | |
241 | } else { | |
242 | start = dbstart; | |
243 | } | |
244 | dbend = ((db->db_blkid + 1) << epbs) - 1; | |
245 | end = (blkid + nblks - 1) >> shift; | |
246 | if (dbend <= end) | |
247 | end = dbend; | |
248 | else if (all) | |
249 | all = trunc; | |
250 | ASSERT3U(start, <=, end); | |
251 | ||
252 | if (db->db_level == 1) { | |
253 | FREE_VERIFY(db, start, end, tx); | |
b128c09f | 254 | blocks_freed = free_blocks(dn, bp, end-start+1, tx); |
34dc7c2f | 255 | arc_buf_freeze(db->db_buf); |
b128c09f BB |
256 | ASSERT(all || blocks_freed == 0 || db->db_last_dirty); |
257 | return (all ? ALL : blocks_freed); | |
34dc7c2f BB |
258 | } |
259 | ||
260 | for (i = start; i <= end; i++, bp++) { | |
261 | if (BP_IS_HOLE(bp)) | |
262 | continue; | |
263 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
264 | err = dbuf_hold_impl(dn, db->db_level-1, i, TRUE, FTAG, &subdb); | |
265 | ASSERT3U(err, ==, 0); | |
266 | rw_exit(&dn->dn_struct_rwlock); | |
267 | ||
b128c09f | 268 | if (free_children(subdb, blkid, nblks, trunc, tx) == ALL) { |
34dc7c2f | 269 | ASSERT3P(subdb->db_blkptr, ==, bp); |
b128c09f | 270 | blocks_freed += free_blocks(dn, bp, 1, tx); |
34dc7c2f BB |
271 | } else { |
272 | all = FALSE; | |
273 | } | |
274 | dbuf_rele(subdb, FTAG); | |
275 | } | |
276 | arc_buf_freeze(db->db_buf); | |
277 | #ifdef ZFS_DEBUG | |
278 | bp -= (end-start)+1; | |
279 | for (i = start; i <= end; i++, bp++) { | |
280 | if (i == start && blkid != 0) | |
281 | continue; | |
282 | else if (i == end && !trunc) | |
283 | continue; | |
284 | ASSERT3U(bp->blk_birth, ==, 0); | |
285 | } | |
286 | #endif | |
b128c09f BB |
287 | ASSERT(all || blocks_freed == 0 || db->db_last_dirty); |
288 | return (all ? ALL : blocks_freed); | |
34dc7c2f BB |
289 | } |
290 | ||
291 | /* | |
292 | * free_range: Traverse the indicated range of the provided file | |
293 | * and "free" all the blocks contained there. | |
294 | */ | |
295 | static void | |
296 | dnode_sync_free_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) | |
297 | { | |
298 | blkptr_t *bp = dn->dn_phys->dn_blkptr; | |
299 | dmu_buf_impl_t *db; | |
300 | int trunc, start, end, shift, i, err; | |
301 | int dnlevel = dn->dn_phys->dn_nlevels; | |
302 | ||
303 | if (blkid > dn->dn_phys->dn_maxblkid) | |
304 | return; | |
305 | ||
306 | ASSERT(dn->dn_phys->dn_maxblkid < UINT64_MAX); | |
307 | trunc = blkid + nblks > dn->dn_phys->dn_maxblkid; | |
308 | if (trunc) | |
309 | nblks = dn->dn_phys->dn_maxblkid - blkid + 1; | |
310 | ||
311 | /* There are no indirect blocks in the object */ | |
312 | if (dnlevel == 1) { | |
313 | if (blkid >= dn->dn_phys->dn_nblkptr) { | |
314 | /* this range was never made persistent */ | |
315 | return; | |
316 | } | |
317 | ASSERT3U(blkid + nblks, <=, dn->dn_phys->dn_nblkptr); | |
b128c09f | 318 | (void) free_blocks(dn, bp + blkid, nblks, tx); |
34dc7c2f BB |
319 | if (trunc) { |
320 | uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * | |
321 | (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
322 | dn->dn_phys->dn_maxblkid = (blkid ? blkid - 1 : 0); | |
323 | ASSERT(off < dn->dn_phys->dn_maxblkid || | |
324 | dn->dn_phys->dn_maxblkid == 0 || | |
b128c09f | 325 | dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); |
34dc7c2f BB |
326 | } |
327 | return; | |
328 | } | |
329 | ||
330 | shift = (dnlevel - 1) * (dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT); | |
331 | start = blkid >> shift; | |
332 | ASSERT(start < dn->dn_phys->dn_nblkptr); | |
333 | end = (blkid + nblks - 1) >> shift; | |
334 | bp += start; | |
335 | for (i = start; i <= end; i++, bp++) { | |
336 | if (BP_IS_HOLE(bp)) | |
337 | continue; | |
338 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
339 | err = dbuf_hold_impl(dn, dnlevel-1, i, TRUE, FTAG, &db); | |
340 | ASSERT3U(err, ==, 0); | |
341 | rw_exit(&dn->dn_struct_rwlock); | |
342 | ||
b128c09f | 343 | if (free_children(db, blkid, nblks, trunc, tx) == ALL) { |
34dc7c2f | 344 | ASSERT3P(db->db_blkptr, ==, bp); |
b128c09f | 345 | (void) free_blocks(dn, bp, 1, tx); |
34dc7c2f BB |
346 | } |
347 | dbuf_rele(db, FTAG); | |
348 | } | |
349 | if (trunc) { | |
350 | uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * | |
351 | (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
352 | dn->dn_phys->dn_maxblkid = (blkid ? blkid - 1 : 0); | |
353 | ASSERT(off < dn->dn_phys->dn_maxblkid || | |
354 | dn->dn_phys->dn_maxblkid == 0 || | |
b128c09f | 355 | dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); |
34dc7c2f BB |
356 | } |
357 | } | |
358 | ||
359 | /* | |
360 | * Try to kick all the dnodes dbufs out of the cache... | |
361 | */ | |
362 | void | |
363 | dnode_evict_dbufs(dnode_t *dn) | |
364 | { | |
365 | int progress; | |
366 | int pass = 0; | |
367 | ||
368 | do { | |
369 | dmu_buf_impl_t *db, marker; | |
370 | int evicting = FALSE; | |
371 | ||
372 | progress = FALSE; | |
373 | mutex_enter(&dn->dn_dbufs_mtx); | |
374 | list_insert_tail(&dn->dn_dbufs, &marker); | |
375 | db = list_head(&dn->dn_dbufs); | |
376 | for (; db != ▮ db = list_head(&dn->dn_dbufs)) { | |
377 | list_remove(&dn->dn_dbufs, db); | |
378 | list_insert_tail(&dn->dn_dbufs, db); | |
379 | ASSERT3P(db->db_dnode, ==, dn); | |
380 | ||
381 | mutex_enter(&db->db_mtx); | |
382 | if (db->db_state == DB_EVICTING) { | |
383 | progress = TRUE; | |
384 | evicting = TRUE; | |
385 | mutex_exit(&db->db_mtx); | |
386 | } else if (refcount_is_zero(&db->db_holds)) { | |
387 | progress = TRUE; | |
34dc7c2f BB |
388 | dbuf_clear(db); /* exits db_mtx for us */ |
389 | } else { | |
390 | mutex_exit(&db->db_mtx); | |
391 | } | |
392 | ||
393 | } | |
394 | list_remove(&dn->dn_dbufs, &marker); | |
395 | /* | |
396 | * NB: we need to drop dn_dbufs_mtx between passes so | |
397 | * that any DB_EVICTING dbufs can make progress. | |
398 | * Ideally, we would have some cv we could wait on, but | |
399 | * since we don't, just wait a bit to give the other | |
400 | * thread a chance to run. | |
401 | */ | |
402 | mutex_exit(&dn->dn_dbufs_mtx); | |
403 | if (evicting) | |
404 | delay(1); | |
405 | pass++; | |
406 | ASSERT(pass < 100); /* sanity check */ | |
407 | } while (progress); | |
408 | ||
409 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
410 | if (dn->dn_bonus && refcount_is_zero(&dn->dn_bonus->db_holds)) { | |
411 | mutex_enter(&dn->dn_bonus->db_mtx); | |
412 | dbuf_evict(dn->dn_bonus); | |
413 | dn->dn_bonus = NULL; | |
414 | } | |
415 | rw_exit(&dn->dn_struct_rwlock); | |
416 | } | |
417 | ||
418 | static void | |
419 | dnode_undirty_dbufs(list_t *list) | |
420 | { | |
421 | dbuf_dirty_record_t *dr; | |
422 | ||
423 | while (dr = list_head(list)) { | |
424 | dmu_buf_impl_t *db = dr->dr_dbuf; | |
425 | uint64_t txg = dr->dr_txg; | |
426 | ||
427 | mutex_enter(&db->db_mtx); | |
428 | /* XXX - use dbuf_undirty()? */ | |
429 | list_remove(list, dr); | |
430 | ASSERT(db->db_last_dirty == dr); | |
431 | db->db_last_dirty = NULL; | |
432 | db->db_dirtycnt -= 1; | |
433 | if (db->db_level == 0) { | |
434 | ASSERT(db->db_blkid == DB_BONUS_BLKID || | |
435 | dr->dt.dl.dr_data == db->db_buf); | |
436 | dbuf_unoverride(dr); | |
437 | mutex_exit(&db->db_mtx); | |
438 | } else { | |
439 | mutex_exit(&db->db_mtx); | |
440 | dnode_undirty_dbufs(&dr->dt.di.dr_children); | |
441 | } | |
442 | kmem_free(dr, sizeof (dbuf_dirty_record_t)); | |
443 | dbuf_rele(db, (void *)(uintptr_t)txg); | |
444 | } | |
445 | } | |
446 | ||
447 | static void | |
448 | dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) | |
449 | { | |
450 | int txgoff = tx->tx_txg & TXG_MASK; | |
451 | ||
452 | ASSERT(dmu_tx_is_syncing(tx)); | |
453 | ||
b128c09f BB |
454 | /* |
455 | * Our contents should have been freed in dnode_sync() by the | |
456 | * free range record inserted by the caller of dnode_free(). | |
457 | */ | |
458 | ASSERT3U(DN_USED_BYTES(dn->dn_phys), ==, 0); | |
459 | ASSERT(BP_IS_HOLE(dn->dn_phys->dn_blkptr)); | |
460 | ||
34dc7c2f BB |
461 | dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); |
462 | dnode_evict_dbufs(dn); | |
463 | ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); | |
464 | ||
465 | /* | |
466 | * XXX - It would be nice to assert this, but we may still | |
467 | * have residual holds from async evictions from the arc... | |
468 | * | |
469 | * zfs_obj_to_path() also depends on this being | |
470 | * commented out. | |
471 | * | |
472 | * ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); | |
473 | */ | |
474 | ||
475 | /* Undirty next bits */ | |
476 | dn->dn_next_nlevels[txgoff] = 0; | |
477 | dn->dn_next_indblkshift[txgoff] = 0; | |
478 | dn->dn_next_blksz[txgoff] = 0; | |
479 | ||
34dc7c2f BB |
480 | /* ASSERT(blkptrs are zero); */ |
481 | ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); | |
482 | ASSERT(dn->dn_type != DMU_OT_NONE); | |
483 | ||
484 | ASSERT(dn->dn_free_txg > 0); | |
485 | if (dn->dn_allocated_txg != dn->dn_free_txg) | |
486 | dbuf_will_dirty(dn->dn_dbuf, tx); | |
487 | bzero(dn->dn_phys, sizeof (dnode_phys_t)); | |
488 | ||
489 | mutex_enter(&dn->dn_mtx); | |
490 | dn->dn_type = DMU_OT_NONE; | |
491 | dn->dn_maxblkid = 0; | |
492 | dn->dn_allocated_txg = 0; | |
493 | dn->dn_free_txg = 0; | |
494 | mutex_exit(&dn->dn_mtx); | |
495 | ||
496 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); | |
497 | ||
498 | dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); | |
499 | /* | |
500 | * Now that we've released our hold, the dnode may | |
501 | * be evicted, so we musn't access it. | |
502 | */ | |
503 | } | |
504 | ||
505 | /* | |
506 | * Write out the dnode's dirty buffers. | |
34dc7c2f BB |
507 | */ |
508 | void | |
509 | dnode_sync(dnode_t *dn, dmu_tx_t *tx) | |
510 | { | |
511 | free_range_t *rp; | |
512 | dnode_phys_t *dnp = dn->dn_phys; | |
513 | int txgoff = tx->tx_txg & TXG_MASK; | |
514 | list_t *list = &dn->dn_dirty_records[txgoff]; | |
9babb374 | 515 | static const dnode_phys_t zerodn = { 0 }; |
34dc7c2f BB |
516 | |
517 | ASSERT(dmu_tx_is_syncing(tx)); | |
518 | ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg); | |
9babb374 BB |
519 | ASSERT(dnp->dn_type != DMU_OT_NONE || |
520 | bcmp(dnp, &zerodn, DNODE_SIZE) == 0); | |
34dc7c2f BB |
521 | DNODE_VERIFY(dn); |
522 | ||
523 | ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf)); | |
524 | ||
9babb374 BB |
525 | if (dmu_objset_userused_enabled(dn->dn_objset) && |
526 | !DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { | |
527 | ASSERT(dn->dn_oldphys == NULL); | |
528 | dn->dn_oldphys = zio_buf_alloc(sizeof (dnode_phys_t)); | |
529 | *dn->dn_oldphys = *dn->dn_phys; /* struct assignment */ | |
530 | dn->dn_phys->dn_flags |= DNODE_FLAG_USERUSED_ACCOUNTED; | |
531 | } else { | |
532 | /* Once we account for it, we should always account for it. */ | |
533 | ASSERT(!(dn->dn_phys->dn_flags & | |
534 | DNODE_FLAG_USERUSED_ACCOUNTED)); | |
535 | } | |
536 | ||
34dc7c2f BB |
537 | mutex_enter(&dn->dn_mtx); |
538 | if (dn->dn_allocated_txg == tx->tx_txg) { | |
539 | /* The dnode is newly allocated or reallocated */ | |
540 | if (dnp->dn_type == DMU_OT_NONE) { | |
541 | /* this is a first alloc, not a realloc */ | |
34dc7c2f | 542 | dnp->dn_nlevels = 1; |
d164b209 | 543 | dnp->dn_nblkptr = dn->dn_nblkptr; |
34dc7c2f BB |
544 | } |
545 | ||
34dc7c2f BB |
546 | dnp->dn_type = dn->dn_type; |
547 | dnp->dn_bonustype = dn->dn_bonustype; | |
548 | dnp->dn_bonuslen = dn->dn_bonuslen; | |
34dc7c2f BB |
549 | } |
550 | ||
551 | ASSERT(dnp->dn_nlevels > 1 || | |
552 | BP_IS_HOLE(&dnp->dn_blkptr[0]) || | |
553 | BP_GET_LSIZE(&dnp->dn_blkptr[0]) == | |
554 | dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
555 | ||
556 | if (dn->dn_next_blksz[txgoff]) { | |
557 | ASSERT(P2PHASE(dn->dn_next_blksz[txgoff], | |
558 | SPA_MINBLOCKSIZE) == 0); | |
559 | ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[0]) || | |
b128c09f | 560 | dn->dn_maxblkid == 0 || list_head(list) != NULL || |
34dc7c2f BB |
561 | dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT == |
562 | dnp->dn_datablkszsec); | |
563 | dnp->dn_datablkszsec = | |
564 | dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT; | |
565 | dn->dn_next_blksz[txgoff] = 0; | |
566 | } | |
567 | ||
568 | if (dn->dn_next_bonuslen[txgoff]) { | |
569 | if (dn->dn_next_bonuslen[txgoff] == DN_ZERO_BONUSLEN) | |
570 | dnp->dn_bonuslen = 0; | |
571 | else | |
572 | dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff]; | |
573 | ASSERT(dnp->dn_bonuslen <= DN_MAX_BONUSLEN); | |
574 | dn->dn_next_bonuslen[txgoff] = 0; | |
575 | } | |
576 | ||
577 | if (dn->dn_next_indblkshift[txgoff]) { | |
578 | ASSERT(dnp->dn_nlevels == 1); | |
579 | dnp->dn_indblkshift = dn->dn_next_indblkshift[txgoff]; | |
580 | dn->dn_next_indblkshift[txgoff] = 0; | |
581 | } | |
582 | ||
583 | /* | |
584 | * Just take the live (open-context) values for checksum and compress. | |
585 | * Strictly speaking it's a future leak, but nothing bad happens if we | |
586 | * start using the new checksum or compress algorithm a little early. | |
587 | */ | |
588 | dnp->dn_checksum = dn->dn_checksum; | |
589 | dnp->dn_compress = dn->dn_compress; | |
590 | ||
591 | mutex_exit(&dn->dn_mtx); | |
592 | ||
593 | /* process all the "freed" ranges in the file */ | |
b128c09f BB |
594 | while (rp = avl_last(&dn->dn_ranges[txgoff])) { |
595 | dnode_sync_free_range(dn, rp->fr_blkid, rp->fr_nblks, tx); | |
596 | /* grab the mutex so we don't race with dnode_block_freed() */ | |
597 | mutex_enter(&dn->dn_mtx); | |
598 | avl_remove(&dn->dn_ranges[txgoff], rp); | |
599 | mutex_exit(&dn->dn_mtx); | |
600 | kmem_free(rp, sizeof (free_range_t)); | |
34dc7c2f | 601 | } |
34dc7c2f | 602 | |
34dc7c2f BB |
603 | if (dn->dn_free_txg > 0 && dn->dn_free_txg <= tx->tx_txg) { |
604 | dnode_sync_free(dn, tx); | |
605 | return; | |
606 | } | |
607 | ||
d164b209 BB |
608 | if (dn->dn_next_nblkptr[txgoff]) { |
609 | /* this should only happen on a realloc */ | |
610 | ASSERT(dn->dn_allocated_txg == tx->tx_txg); | |
611 | if (dn->dn_next_nblkptr[txgoff] > dnp->dn_nblkptr) { | |
612 | /* zero the new blkptrs we are gaining */ | |
613 | bzero(dnp->dn_blkptr + dnp->dn_nblkptr, | |
614 | sizeof (blkptr_t) * | |
615 | (dn->dn_next_nblkptr[txgoff] - dnp->dn_nblkptr)); | |
616 | #ifdef ZFS_DEBUG | |
617 | } else { | |
618 | int i; | |
619 | ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr); | |
620 | /* the blkptrs we are losing better be unallocated */ | |
621 | for (i = dn->dn_next_nblkptr[txgoff]; | |
622 | i < dnp->dn_nblkptr; i++) | |
623 | ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i])); | |
624 | #endif | |
625 | } | |
626 | mutex_enter(&dn->dn_mtx); | |
627 | dnp->dn_nblkptr = dn->dn_next_nblkptr[txgoff]; | |
628 | dn->dn_next_nblkptr[txgoff] = 0; | |
629 | mutex_exit(&dn->dn_mtx); | |
630 | } | |
631 | ||
34dc7c2f BB |
632 | if (dn->dn_next_nlevels[txgoff]) { |
633 | dnode_increase_indirection(dn, tx); | |
634 | dn->dn_next_nlevels[txgoff] = 0; | |
635 | } | |
636 | ||
637 | dbuf_sync_list(list, tx); | |
638 | ||
9babb374 | 639 | if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { |
34dc7c2f BB |
640 | ASSERT3P(list_head(list), ==, NULL); |
641 | dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); | |
642 | } | |
643 | ||
644 | /* | |
645 | * Although we have dropped our reference to the dnode, it | |
646 | * can't be evicted until its written, and we haven't yet | |
647 | * initiated the IO for the dnode's dbuf. | |
648 | */ | |
649 | } |