]>
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 | */ | |
9ae529ec | 21 | |
34dc7c2f | 22 | /* |
428870ff | 23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
fa86b5db | 24 | * Copyright (c) 2013 by Delphix. All rights reserved. |
34dc7c2f BB |
25 | */ |
26 | ||
34dc7c2f BB |
27 | #include <sys/zfs_context.h> |
28 | #include <sys/dbuf.h> | |
29 | #include <sys/dnode.h> | |
30 | #include <sys/dmu.h> | |
31 | #include <sys/dmu_tx.h> | |
32 | #include <sys/dmu_objset.h> | |
33 | #include <sys/dsl_dataset.h> | |
34 | #include <sys/spa.h> | |
35 | ||
36 | static void | |
37 | dnode_increase_indirection(dnode_t *dn, dmu_tx_t *tx) | |
38 | { | |
39 | dmu_buf_impl_t *db; | |
40 | int txgoff = tx->tx_txg & TXG_MASK; | |
41 | int nblkptr = dn->dn_phys->dn_nblkptr; | |
42 | int old_toplvl = dn->dn_phys->dn_nlevels - 1; | |
43 | int new_level = dn->dn_next_nlevels[txgoff]; | |
44 | int i; | |
45 | ||
46 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
47 | ||
48 | /* this dnode can't be paged out because it's dirty */ | |
49 | ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); | |
50 | ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); | |
51 | ASSERT(new_level > 1 && dn->dn_phys->dn_nlevels > 0); | |
52 | ||
53 | db = dbuf_hold_level(dn, dn->dn_phys->dn_nlevels, 0, FTAG); | |
54 | ASSERT(db != NULL); | |
55 | ||
56 | dn->dn_phys->dn_nlevels = new_level; | |
57 | dprintf("os=%p obj=%llu, increase to %d\n", dn->dn_objset, | |
58 | dn->dn_object, dn->dn_phys->dn_nlevels); | |
59 | ||
60 | /* check for existing blkptrs in the dnode */ | |
61 | for (i = 0; i < nblkptr; i++) | |
62 | if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[i])) | |
63 | break; | |
64 | if (i != nblkptr) { | |
65 | /* transfer dnode's block pointers to new indirect block */ | |
66 | (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); | |
67 | ASSERT(db->db.db_data); | |
68 | ASSERT(arc_released(db->db_buf)); | |
69 | ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); | |
70 | bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, | |
71 | sizeof (blkptr_t) * nblkptr); | |
72 | arc_buf_freeze(db->db_buf); | |
73 | } | |
74 | ||
75 | /* set dbuf's parent pointers to new indirect buf */ | |
76 | for (i = 0; i < nblkptr; i++) { | |
77 | dmu_buf_impl_t *child = dbuf_find(dn, old_toplvl, i); | |
78 | ||
79 | if (child == NULL) | |
80 | continue; | |
572e2857 BB |
81 | #ifdef DEBUG |
82 | DB_DNODE_ENTER(child); | |
83 | ASSERT3P(DB_DNODE(child), ==, dn); | |
84 | DB_DNODE_EXIT(child); | |
85 | #endif /* DEBUG */ | |
34dc7c2f BB |
86 | if (child->db_parent && child->db_parent != dn->dn_dbuf) { |
87 | ASSERT(child->db_parent->db_level == db->db_level); | |
88 | ASSERT(child->db_blkptr != | |
89 | &dn->dn_phys->dn_blkptr[child->db_blkid]); | |
90 | mutex_exit(&child->db_mtx); | |
91 | continue; | |
92 | } | |
93 | ASSERT(child->db_parent == NULL || | |
94 | child->db_parent == dn->dn_dbuf); | |
95 | ||
96 | child->db_parent = db; | |
97 | dbuf_add_ref(db, child); | |
98 | if (db->db.db_data) | |
99 | child->db_blkptr = (blkptr_t *)db->db.db_data + i; | |
100 | else | |
101 | child->db_blkptr = NULL; | |
102 | dprintf_dbuf_bp(child, child->db_blkptr, | |
103 | "changed db_blkptr to new indirect %s", ""); | |
104 | ||
105 | mutex_exit(&child->db_mtx); | |
106 | } | |
107 | ||
108 | bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); | |
109 | ||
110 | dbuf_rele(db, FTAG); | |
111 | ||
112 | rw_exit(&dn->dn_struct_rwlock); | |
113 | } | |
114 | ||
b128c09f | 115 | static int |
34dc7c2f BB |
116 | free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) |
117 | { | |
b128c09f | 118 | dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; |
34dc7c2f | 119 | uint64_t bytesfreed = 0; |
b128c09f | 120 | int i, blocks_freed = 0; |
34dc7c2f | 121 | |
b128c09f | 122 | dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num); |
34dc7c2f BB |
123 | |
124 | for (i = 0; i < num; i++, bp++) { | |
125 | if (BP_IS_HOLE(bp)) | |
126 | continue; | |
127 | ||
428870ff | 128 | bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE); |
34dc7c2f | 129 | ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); |
34dc7c2f | 130 | bzero(bp, sizeof (blkptr_t)); |
b128c09f | 131 | blocks_freed += 1; |
34dc7c2f BB |
132 | } |
133 | dnode_diduse_space(dn, -bytesfreed); | |
b128c09f | 134 | return (blocks_freed); |
34dc7c2f BB |
135 | } |
136 | ||
137 | #ifdef ZFS_DEBUG | |
138 | static void | |
139 | free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) | |
140 | { | |
141 | int off, num; | |
142 | int i, err, epbs; | |
143 | uint64_t txg = tx->tx_txg; | |
572e2857 | 144 | dnode_t *dn; |
34dc7c2f | 145 | |
572e2857 BB |
146 | DB_DNODE_ENTER(db); |
147 | dn = DB_DNODE(db); | |
148 | epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
34dc7c2f BB |
149 | off = start - (db->db_blkid * 1<<epbs); |
150 | num = end - start + 1; | |
151 | ||
152 | ASSERT3U(off, >=, 0); | |
153 | ASSERT3U(num, >=, 0); | |
154 | ASSERT3U(db->db_level, >, 0); | |
572e2857 | 155 | ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); |
34dc7c2f BB |
156 | ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); |
157 | ASSERT(db->db_blkptr != NULL); | |
158 | ||
159 | for (i = off; i < off+num; i++) { | |
160 | uint64_t *buf; | |
161 | dmu_buf_impl_t *child; | |
162 | dbuf_dirty_record_t *dr; | |
163 | int j; | |
164 | ||
165 | ASSERT(db->db_level == 1); | |
166 | ||
572e2857 BB |
167 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
168 | err = dbuf_hold_impl(dn, db->db_level-1, | |
34dc7c2f | 169 | (db->db_blkid << epbs) + i, TRUE, FTAG, &child); |
572e2857 | 170 | rw_exit(&dn->dn_struct_rwlock); |
34dc7c2f BB |
171 | if (err == ENOENT) |
172 | continue; | |
173 | ASSERT(err == 0); | |
174 | ASSERT(child->db_level == 0); | |
175 | dr = child->db_last_dirty; | |
176 | while (dr && dr->dr_txg > txg) | |
177 | dr = dr->dr_next; | |
178 | ASSERT(dr == NULL || dr->dr_txg == txg); | |
179 | ||
180 | /* data_old better be zeroed */ | |
181 | if (dr) { | |
182 | buf = dr->dt.dl.dr_data->b_data; | |
183 | for (j = 0; j < child->db.db_size >> 3; j++) { | |
184 | if (buf[j] != 0) { | |
185 | panic("freed data not zero: " | |
186 | "child=%p i=%d off=%d num=%d\n", | |
b128c09f | 187 | (void *)child, i, off, num); |
34dc7c2f BB |
188 | } |
189 | } | |
190 | } | |
191 | ||
192 | /* | |
193 | * db_data better be zeroed unless it's dirty in a | |
194 | * future txg. | |
195 | */ | |
196 | mutex_enter(&child->db_mtx); | |
197 | buf = child->db.db_data; | |
198 | if (buf != NULL && child->db_state != DB_FILL && | |
199 | child->db_last_dirty == NULL) { | |
200 | for (j = 0; j < child->db.db_size >> 3; j++) { | |
201 | if (buf[j] != 0) { | |
202 | panic("freed data not zero: " | |
203 | "child=%p i=%d off=%d num=%d\n", | |
b128c09f | 204 | (void *)child, i, off, num); |
34dc7c2f BB |
205 | } |
206 | } | |
207 | } | |
208 | mutex_exit(&child->db_mtx); | |
209 | ||
210 | dbuf_rele(child, FTAG); | |
211 | } | |
572e2857 | 212 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
213 | } |
214 | #endif | |
215 | ||
b128c09f BB |
216 | #define ALL -1 |
217 | ||
34dc7c2f BB |
218 | static int |
219 | free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, int trunc, | |
220 | dmu_tx_t *tx) | |
221 | { | |
572e2857 | 222 | dnode_t *dn; |
34dc7c2f BB |
223 | blkptr_t *bp; |
224 | dmu_buf_impl_t *subdb; | |
225 | uint64_t start, end, dbstart, dbend, i; | |
226 | int epbs, shift, err; | |
227 | int all = TRUE; | |
b128c09f BB |
228 | int blocks_freed = 0; |
229 | ||
230 | /* | |
231 | * There is a small possibility that this block will not be cached: | |
232 | * 1 - if level > 1 and there are no children with level <= 1 | |
233 | * 2 - if we didn't get a dirty hold (because this block had just | |
234 | * finished being written -- and so had no holds), and then this | |
235 | * block got evicted before we got here. | |
236 | */ | |
237 | if (db->db_state != DB_CACHED) | |
238 | (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); | |
34dc7c2f | 239 | |
428870ff | 240 | dbuf_release_bp(db); |
34dc7c2f BB |
241 | bp = (blkptr_t *)db->db.db_data; |
242 | ||
572e2857 BB |
243 | DB_DNODE_ENTER(db); |
244 | dn = DB_DNODE(db); | |
245 | epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
34dc7c2f BB |
246 | shift = (db->db_level - 1) * epbs; |
247 | dbstart = db->db_blkid << epbs; | |
248 | start = blkid >> shift; | |
249 | if (dbstart < start) { | |
250 | bp += start - dbstart; | |
251 | all = FALSE; | |
252 | } else { | |
253 | start = dbstart; | |
254 | } | |
255 | dbend = ((db->db_blkid + 1) << epbs) - 1; | |
256 | end = (blkid + nblks - 1) >> shift; | |
257 | if (dbend <= end) | |
258 | end = dbend; | |
259 | else if (all) | |
260 | all = trunc; | |
261 | ASSERT3U(start, <=, end); | |
262 | ||
263 | if (db->db_level == 1) { | |
264 | FREE_VERIFY(db, start, end, tx); | |
b128c09f | 265 | blocks_freed = free_blocks(dn, bp, end-start+1, tx); |
34dc7c2f | 266 | arc_buf_freeze(db->db_buf); |
b128c09f | 267 | ASSERT(all || blocks_freed == 0 || db->db_last_dirty); |
572e2857 | 268 | DB_DNODE_EXIT(db); |
b128c09f | 269 | return (all ? ALL : blocks_freed); |
34dc7c2f BB |
270 | } |
271 | ||
272 | for (i = start; i <= end; i++, bp++) { | |
273 | if (BP_IS_HOLE(bp)) | |
274 | continue; | |
275 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
276 | err = dbuf_hold_impl(dn, db->db_level-1, i, TRUE, FTAG, &subdb); | |
c99c9001 | 277 | ASSERT0(err); |
34dc7c2f BB |
278 | rw_exit(&dn->dn_struct_rwlock); |
279 | ||
b128c09f | 280 | if (free_children(subdb, blkid, nblks, trunc, tx) == ALL) { |
34dc7c2f | 281 | ASSERT3P(subdb->db_blkptr, ==, bp); |
b128c09f | 282 | blocks_freed += free_blocks(dn, bp, 1, tx); |
34dc7c2f BB |
283 | } else { |
284 | all = FALSE; | |
285 | } | |
286 | dbuf_rele(subdb, FTAG); | |
287 | } | |
572e2857 | 288 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
289 | arc_buf_freeze(db->db_buf); |
290 | #ifdef ZFS_DEBUG | |
291 | bp -= (end-start)+1; | |
292 | for (i = start; i <= end; i++, bp++) { | |
293 | if (i == start && blkid != 0) | |
294 | continue; | |
295 | else if (i == end && !trunc) | |
296 | continue; | |
c99c9001 | 297 | ASSERT0(bp->blk_birth); |
34dc7c2f BB |
298 | } |
299 | #endif | |
b128c09f BB |
300 | ASSERT(all || blocks_freed == 0 || db->db_last_dirty); |
301 | return (all ? ALL : blocks_freed); | |
34dc7c2f BB |
302 | } |
303 | ||
304 | /* | |
d3cc8b15 | 305 | * Traverse the indicated range of the provided file |
34dc7c2f BB |
306 | * and "free" all the blocks contained there. |
307 | */ | |
308 | static void | |
309 | dnode_sync_free_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) | |
310 | { | |
311 | blkptr_t *bp = dn->dn_phys->dn_blkptr; | |
312 | dmu_buf_impl_t *db; | |
313 | int trunc, start, end, shift, i, err; | |
314 | int dnlevel = dn->dn_phys->dn_nlevels; | |
315 | ||
316 | if (blkid > dn->dn_phys->dn_maxblkid) | |
317 | return; | |
318 | ||
319 | ASSERT(dn->dn_phys->dn_maxblkid < UINT64_MAX); | |
320 | trunc = blkid + nblks > dn->dn_phys->dn_maxblkid; | |
321 | if (trunc) | |
322 | nblks = dn->dn_phys->dn_maxblkid - blkid + 1; | |
323 | ||
324 | /* There are no indirect blocks in the object */ | |
325 | if (dnlevel == 1) { | |
326 | if (blkid >= dn->dn_phys->dn_nblkptr) { | |
327 | /* this range was never made persistent */ | |
328 | return; | |
329 | } | |
330 | ASSERT3U(blkid + nblks, <=, dn->dn_phys->dn_nblkptr); | |
b128c09f | 331 | (void) free_blocks(dn, bp + blkid, nblks, tx); |
34dc7c2f | 332 | if (trunc) { |
1fde1e37 BB |
333 | ASSERTV(uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * |
334 | (dn->dn_phys->dn_datablkszsec<<SPA_MINBLOCKSHIFT)); | |
34dc7c2f BB |
335 | dn->dn_phys->dn_maxblkid = (blkid ? blkid - 1 : 0); |
336 | ASSERT(off < dn->dn_phys->dn_maxblkid || | |
337 | dn->dn_phys->dn_maxblkid == 0 || | |
b128c09f | 338 | dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); |
34dc7c2f BB |
339 | } |
340 | return; | |
341 | } | |
342 | ||
343 | shift = (dnlevel - 1) * (dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT); | |
344 | start = blkid >> shift; | |
345 | ASSERT(start < dn->dn_phys->dn_nblkptr); | |
346 | end = (blkid + nblks - 1) >> shift; | |
347 | bp += start; | |
348 | for (i = start; i <= end; i++, bp++) { | |
349 | if (BP_IS_HOLE(bp)) | |
350 | continue; | |
351 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
352 | err = dbuf_hold_impl(dn, dnlevel-1, i, TRUE, FTAG, &db); | |
c99c9001 | 353 | ASSERT0(err); |
34dc7c2f BB |
354 | rw_exit(&dn->dn_struct_rwlock); |
355 | ||
b128c09f | 356 | if (free_children(db, blkid, nblks, trunc, tx) == ALL) { |
34dc7c2f | 357 | ASSERT3P(db->db_blkptr, ==, bp); |
b128c09f | 358 | (void) free_blocks(dn, bp, 1, tx); |
34dc7c2f BB |
359 | } |
360 | dbuf_rele(db, FTAG); | |
361 | } | |
362 | if (trunc) { | |
1fde1e37 BB |
363 | ASSERTV(uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * |
364 | (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT)); | |
34dc7c2f BB |
365 | dn->dn_phys->dn_maxblkid = (blkid ? blkid - 1 : 0); |
366 | ASSERT(off < dn->dn_phys->dn_maxblkid || | |
367 | dn->dn_phys->dn_maxblkid == 0 || | |
b128c09f | 368 | dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); |
34dc7c2f BB |
369 | } |
370 | } | |
371 | ||
372 | /* | |
d3cc8b15 | 373 | * Try to kick all the dnode's dbufs out of the cache... |
34dc7c2f BB |
374 | */ |
375 | void | |
376 | dnode_evict_dbufs(dnode_t *dn) | |
377 | { | |
378 | int progress; | |
379 | int pass = 0; | |
380 | ||
381 | do { | |
382 | dmu_buf_impl_t *db, marker; | |
383 | int evicting = FALSE; | |
384 | ||
385 | progress = FALSE; | |
386 | mutex_enter(&dn->dn_dbufs_mtx); | |
387 | list_insert_tail(&dn->dn_dbufs, &marker); | |
388 | db = list_head(&dn->dn_dbufs); | |
389 | for (; db != ▮ db = list_head(&dn->dn_dbufs)) { | |
390 | list_remove(&dn->dn_dbufs, db); | |
391 | list_insert_tail(&dn->dn_dbufs, db); | |
572e2857 BB |
392 | #ifdef DEBUG |
393 | DB_DNODE_ENTER(db); | |
394 | ASSERT3P(DB_DNODE(db), ==, dn); | |
395 | DB_DNODE_EXIT(db); | |
396 | #endif /* DEBUG */ | |
34dc7c2f BB |
397 | |
398 | mutex_enter(&db->db_mtx); | |
399 | if (db->db_state == DB_EVICTING) { | |
400 | progress = TRUE; | |
401 | evicting = TRUE; | |
402 | mutex_exit(&db->db_mtx); | |
403 | } else if (refcount_is_zero(&db->db_holds)) { | |
404 | progress = TRUE; | |
34dc7c2f BB |
405 | dbuf_clear(db); /* exits db_mtx for us */ |
406 | } else { | |
407 | mutex_exit(&db->db_mtx); | |
408 | } | |
409 | ||
410 | } | |
411 | list_remove(&dn->dn_dbufs, &marker); | |
412 | /* | |
413 | * NB: we need to drop dn_dbufs_mtx between passes so | |
414 | * that any DB_EVICTING dbufs can make progress. | |
415 | * Ideally, we would have some cv we could wait on, but | |
416 | * since we don't, just wait a bit to give the other | |
417 | * thread a chance to run. | |
418 | */ | |
419 | mutex_exit(&dn->dn_dbufs_mtx); | |
420 | if (evicting) | |
421 | delay(1); | |
422 | pass++; | |
754c6663 BB |
423 | if ((pass % 100) == 0) |
424 | dprintf("Exceeded %d passes evicting dbufs\n", pass); | |
34dc7c2f BB |
425 | } while (progress); |
426 | ||
754c6663 BB |
427 | if (pass >= 100) |
428 | dprintf("Required %d passes to evict dbufs\n", pass); | |
429 | ||
34dc7c2f BB |
430 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); |
431 | if (dn->dn_bonus && refcount_is_zero(&dn->dn_bonus->db_holds)) { | |
432 | mutex_enter(&dn->dn_bonus->db_mtx); | |
433 | dbuf_evict(dn->dn_bonus); | |
434 | dn->dn_bonus = NULL; | |
435 | } | |
436 | rw_exit(&dn->dn_struct_rwlock); | |
437 | } | |
438 | ||
439 | static void | |
440 | dnode_undirty_dbufs(list_t *list) | |
441 | { | |
442 | dbuf_dirty_record_t *dr; | |
443 | ||
c65aa5b2 | 444 | while ((dr = list_head(list))) { |
34dc7c2f BB |
445 | dmu_buf_impl_t *db = dr->dr_dbuf; |
446 | uint64_t txg = dr->dr_txg; | |
447 | ||
428870ff BB |
448 | if (db->db_level != 0) |
449 | dnode_undirty_dbufs(&dr->dt.di.dr_children); | |
450 | ||
34dc7c2f BB |
451 | mutex_enter(&db->db_mtx); |
452 | /* XXX - use dbuf_undirty()? */ | |
453 | list_remove(list, dr); | |
454 | ASSERT(db->db_last_dirty == dr); | |
455 | db->db_last_dirty = NULL; | |
456 | db->db_dirtycnt -= 1; | |
457 | if (db->db_level == 0) { | |
428870ff | 458 | ASSERT(db->db_blkid == DMU_BONUS_BLKID || |
34dc7c2f BB |
459 | dr->dt.dl.dr_data == db->db_buf); |
460 | dbuf_unoverride(dr); | |
34dc7c2f BB |
461 | } |
462 | kmem_free(dr, sizeof (dbuf_dirty_record_t)); | |
428870ff | 463 | dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); |
34dc7c2f BB |
464 | } |
465 | } | |
466 | ||
467 | static void | |
468 | dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) | |
469 | { | |
470 | int txgoff = tx->tx_txg & TXG_MASK; | |
471 | ||
472 | ASSERT(dmu_tx_is_syncing(tx)); | |
473 | ||
b128c09f BB |
474 | /* |
475 | * Our contents should have been freed in dnode_sync() by the | |
476 | * free range record inserted by the caller of dnode_free(). | |
477 | */ | |
c99c9001 | 478 | ASSERT0(DN_USED_BYTES(dn->dn_phys)); |
b128c09f BB |
479 | ASSERT(BP_IS_HOLE(dn->dn_phys->dn_blkptr)); |
480 | ||
34dc7c2f BB |
481 | dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); |
482 | dnode_evict_dbufs(dn); | |
483 | ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); | |
13fe0198 | 484 | ASSERT3P(dn->dn_bonus, ==, NULL); |
34dc7c2f BB |
485 | |
486 | /* | |
487 | * XXX - It would be nice to assert this, but we may still | |
488 | * have residual holds from async evictions from the arc... | |
489 | * | |
490 | * zfs_obj_to_path() also depends on this being | |
491 | * commented out. | |
492 | * | |
493 | * ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); | |
494 | */ | |
495 | ||
496 | /* Undirty next bits */ | |
497 | dn->dn_next_nlevels[txgoff] = 0; | |
498 | dn->dn_next_indblkshift[txgoff] = 0; | |
499 | dn->dn_next_blksz[txgoff] = 0; | |
500 | ||
34dc7c2f BB |
501 | /* ASSERT(blkptrs are zero); */ |
502 | ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); | |
503 | ASSERT(dn->dn_type != DMU_OT_NONE); | |
504 | ||
505 | ASSERT(dn->dn_free_txg > 0); | |
506 | if (dn->dn_allocated_txg != dn->dn_free_txg) | |
507 | dbuf_will_dirty(dn->dn_dbuf, tx); | |
508 | bzero(dn->dn_phys, sizeof (dnode_phys_t)); | |
509 | ||
510 | mutex_enter(&dn->dn_mtx); | |
511 | dn->dn_type = DMU_OT_NONE; | |
512 | dn->dn_maxblkid = 0; | |
513 | dn->dn_allocated_txg = 0; | |
514 | dn->dn_free_txg = 0; | |
428870ff | 515 | dn->dn_have_spill = B_FALSE; |
34dc7c2f BB |
516 | mutex_exit(&dn->dn_mtx); |
517 | ||
518 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); | |
519 | ||
520 | dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); | |
521 | /* | |
522 | * Now that we've released our hold, the dnode may | |
523 | * be evicted, so we musn't access it. | |
524 | */ | |
525 | } | |
526 | ||
527 | /* | |
528 | * Write out the dnode's dirty buffers. | |
34dc7c2f BB |
529 | */ |
530 | void | |
531 | dnode_sync(dnode_t *dn, dmu_tx_t *tx) | |
532 | { | |
533 | free_range_t *rp; | |
534 | dnode_phys_t *dnp = dn->dn_phys; | |
535 | int txgoff = tx->tx_txg & TXG_MASK; | |
536 | list_t *list = &dn->dn_dirty_records[txgoff]; | |
428870ff | 537 | boolean_t kill_spill = B_FALSE; |
1fde1e37 | 538 | ASSERTV(static const dnode_phys_t zerodn = { 0 }); |
34dc7c2f BB |
539 | |
540 | ASSERT(dmu_tx_is_syncing(tx)); | |
541 | ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg); | |
9babb374 BB |
542 | ASSERT(dnp->dn_type != DMU_OT_NONE || |
543 | bcmp(dnp, &zerodn, DNODE_SIZE) == 0); | |
34dc7c2f BB |
544 | DNODE_VERIFY(dn); |
545 | ||
546 | ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf)); | |
547 | ||
9babb374 BB |
548 | if (dmu_objset_userused_enabled(dn->dn_objset) && |
549 | !DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { | |
428870ff BB |
550 | mutex_enter(&dn->dn_mtx); |
551 | dn->dn_oldused = DN_USED_BYTES(dn->dn_phys); | |
552 | dn->dn_oldflags = dn->dn_phys->dn_flags; | |
9babb374 | 553 | dn->dn_phys->dn_flags |= DNODE_FLAG_USERUSED_ACCOUNTED; |
428870ff BB |
554 | mutex_exit(&dn->dn_mtx); |
555 | dmu_objset_userquota_get_ids(dn, B_FALSE, tx); | |
9babb374 BB |
556 | } else { |
557 | /* Once we account for it, we should always account for it. */ | |
558 | ASSERT(!(dn->dn_phys->dn_flags & | |
559 | DNODE_FLAG_USERUSED_ACCOUNTED)); | |
560 | } | |
561 | ||
34dc7c2f BB |
562 | mutex_enter(&dn->dn_mtx); |
563 | if (dn->dn_allocated_txg == tx->tx_txg) { | |
564 | /* The dnode is newly allocated or reallocated */ | |
565 | if (dnp->dn_type == DMU_OT_NONE) { | |
566 | /* this is a first alloc, not a realloc */ | |
34dc7c2f | 567 | dnp->dn_nlevels = 1; |
d164b209 | 568 | dnp->dn_nblkptr = dn->dn_nblkptr; |
34dc7c2f BB |
569 | } |
570 | ||
34dc7c2f BB |
571 | dnp->dn_type = dn->dn_type; |
572 | dnp->dn_bonustype = dn->dn_bonustype; | |
573 | dnp->dn_bonuslen = dn->dn_bonuslen; | |
34dc7c2f BB |
574 | } |
575 | ||
576 | ASSERT(dnp->dn_nlevels > 1 || | |
577 | BP_IS_HOLE(&dnp->dn_blkptr[0]) || | |
578 | BP_GET_LSIZE(&dnp->dn_blkptr[0]) == | |
579 | dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
580 | ||
fa86b5db MA |
581 | if (dn->dn_next_type[txgoff] != 0) { |
582 | dnp->dn_type = dn->dn_type; | |
583 | dn->dn_next_type[txgoff] = 0; | |
584 | } | |
585 | ||
586 | if (dn->dn_next_blksz[txgoff] != 0) { | |
34dc7c2f BB |
587 | ASSERT(P2PHASE(dn->dn_next_blksz[txgoff], |
588 | SPA_MINBLOCKSIZE) == 0); | |
589 | ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[0]) || | |
b128c09f | 590 | dn->dn_maxblkid == 0 || list_head(list) != NULL || |
428870ff | 591 | avl_last(&dn->dn_ranges[txgoff]) || |
34dc7c2f BB |
592 | dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT == |
593 | dnp->dn_datablkszsec); | |
594 | dnp->dn_datablkszsec = | |
595 | dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT; | |
596 | dn->dn_next_blksz[txgoff] = 0; | |
597 | } | |
598 | ||
fa86b5db | 599 | if (dn->dn_next_bonuslen[txgoff] != 0) { |
34dc7c2f BB |
600 | if (dn->dn_next_bonuslen[txgoff] == DN_ZERO_BONUSLEN) |
601 | dnp->dn_bonuslen = 0; | |
602 | else | |
603 | dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff]; | |
604 | ASSERT(dnp->dn_bonuslen <= DN_MAX_BONUSLEN); | |
605 | dn->dn_next_bonuslen[txgoff] = 0; | |
606 | } | |
607 | ||
fa86b5db | 608 | if (dn->dn_next_bonustype[txgoff] != 0) { |
9ae529ec | 609 | ASSERT(DMU_OT_IS_VALID(dn->dn_next_bonustype[txgoff])); |
428870ff BB |
610 | dnp->dn_bonustype = dn->dn_next_bonustype[txgoff]; |
611 | dn->dn_next_bonustype[txgoff] = 0; | |
612 | } | |
613 | ||
614 | /* | |
615 | * We will either remove a spill block when a file is being removed | |
616 | * or we have been asked to remove it. | |
617 | */ | |
618 | if (dn->dn_rm_spillblk[txgoff] || | |
619 | ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) && | |
620 | dn->dn_free_txg > 0 && dn->dn_free_txg <= tx->tx_txg)) { | |
621 | if ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) | |
622 | kill_spill = B_TRUE; | |
623 | dn->dn_rm_spillblk[txgoff] = 0; | |
624 | } | |
625 | ||
fa86b5db | 626 | if (dn->dn_next_indblkshift[txgoff] != 0) { |
34dc7c2f BB |
627 | ASSERT(dnp->dn_nlevels == 1); |
628 | dnp->dn_indblkshift = dn->dn_next_indblkshift[txgoff]; | |
629 | dn->dn_next_indblkshift[txgoff] = 0; | |
630 | } | |
631 | ||
632 | /* | |
633 | * Just take the live (open-context) values for checksum and compress. | |
634 | * Strictly speaking it's a future leak, but nothing bad happens if we | |
635 | * start using the new checksum or compress algorithm a little early. | |
636 | */ | |
637 | dnp->dn_checksum = dn->dn_checksum; | |
638 | dnp->dn_compress = dn->dn_compress; | |
639 | ||
640 | mutex_exit(&dn->dn_mtx); | |
641 | ||
428870ff BB |
642 | if (kill_spill) { |
643 | (void) free_blocks(dn, &dn->dn_phys->dn_spill, 1, tx); | |
644 | mutex_enter(&dn->dn_mtx); | |
645 | dnp->dn_flags &= ~DNODE_FLAG_SPILL_BLKPTR; | |
646 | mutex_exit(&dn->dn_mtx); | |
647 | } | |
648 | ||
34dc7c2f | 649 | /* process all the "freed" ranges in the file */ |
c65aa5b2 | 650 | while ((rp = avl_last(&dn->dn_ranges[txgoff]))) { |
b128c09f BB |
651 | dnode_sync_free_range(dn, rp->fr_blkid, rp->fr_nblks, tx); |
652 | /* grab the mutex so we don't race with dnode_block_freed() */ | |
653 | mutex_enter(&dn->dn_mtx); | |
654 | avl_remove(&dn->dn_ranges[txgoff], rp); | |
655 | mutex_exit(&dn->dn_mtx); | |
656 | kmem_free(rp, sizeof (free_range_t)); | |
34dc7c2f | 657 | } |
34dc7c2f | 658 | |
34dc7c2f BB |
659 | if (dn->dn_free_txg > 0 && dn->dn_free_txg <= tx->tx_txg) { |
660 | dnode_sync_free(dn, tx); | |
661 | return; | |
662 | } | |
663 | ||
d164b209 BB |
664 | if (dn->dn_next_nblkptr[txgoff]) { |
665 | /* this should only happen on a realloc */ | |
666 | ASSERT(dn->dn_allocated_txg == tx->tx_txg); | |
667 | if (dn->dn_next_nblkptr[txgoff] > dnp->dn_nblkptr) { | |
668 | /* zero the new blkptrs we are gaining */ | |
669 | bzero(dnp->dn_blkptr + dnp->dn_nblkptr, | |
670 | sizeof (blkptr_t) * | |
671 | (dn->dn_next_nblkptr[txgoff] - dnp->dn_nblkptr)); | |
672 | #ifdef ZFS_DEBUG | |
673 | } else { | |
674 | int i; | |
675 | ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr); | |
676 | /* the blkptrs we are losing better be unallocated */ | |
37f000c5 NB |
677 | for (i = 0; i < dnp->dn_nblkptr; i++) { |
678 | if (i >= dn->dn_next_nblkptr[txgoff]) | |
679 | ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i])); | |
680 | } | |
d164b209 BB |
681 | #endif |
682 | } | |
683 | mutex_enter(&dn->dn_mtx); | |
684 | dnp->dn_nblkptr = dn->dn_next_nblkptr[txgoff]; | |
685 | dn->dn_next_nblkptr[txgoff] = 0; | |
686 | mutex_exit(&dn->dn_mtx); | |
687 | } | |
688 | ||
34dc7c2f BB |
689 | if (dn->dn_next_nlevels[txgoff]) { |
690 | dnode_increase_indirection(dn, tx); | |
691 | dn->dn_next_nlevels[txgoff] = 0; | |
692 | } | |
693 | ||
694 | dbuf_sync_list(list, tx); | |
695 | ||
9babb374 | 696 | if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { |
34dc7c2f BB |
697 | ASSERT3P(list_head(list), ==, NULL); |
698 | dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); | |
699 | } | |
700 | ||
701 | /* | |
702 | * Although we have dropped our reference to the dnode, it | |
703 | * can't be evicted until its written, and we haven't yet | |
704 | * initiated the IO for the dnode's dbuf. | |
705 | */ | |
706 | } |