]>
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. |
01937958 | 24 | * Copyright (c) 2012, 2018 by Delphix. All rights reserved. |
0c66c32d | 25 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. |
34dc7c2f BB |
26 | */ |
27 | ||
34dc7c2f BB |
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> | |
9bd274dd | 36 | #include <sys/range_tree.h> |
b0bc7a84 | 37 | #include <sys/zfeature.h> |
34dc7c2f BB |
38 | |
39 | static void | |
40 | dnode_increase_indirection(dnode_t *dn, dmu_tx_t *tx) | |
41 | { | |
42 | dmu_buf_impl_t *db; | |
43 | int txgoff = tx->tx_txg & TXG_MASK; | |
44 | int nblkptr = dn->dn_phys->dn_nblkptr; | |
45 | int old_toplvl = dn->dn_phys->dn_nlevels - 1; | |
46 | int new_level = dn->dn_next_nlevels[txgoff]; | |
47 | int i; | |
48 | ||
49 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
50 | ||
51 | /* this dnode can't be paged out because it's dirty */ | |
52 | ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); | |
53 | ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); | |
54 | ASSERT(new_level > 1 && dn->dn_phys->dn_nlevels > 0); | |
55 | ||
56 | db = dbuf_hold_level(dn, dn->dn_phys->dn_nlevels, 0, FTAG); | |
57 | ASSERT(db != NULL); | |
58 | ||
59 | dn->dn_phys->dn_nlevels = new_level; | |
60 | dprintf("os=%p obj=%llu, increase to %d\n", dn->dn_objset, | |
61 | dn->dn_object, dn->dn_phys->dn_nlevels); | |
62 | ||
bc77ba73 PD |
63 | /* transfer dnode's block pointers to new indirect block */ |
64 | (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); | |
65 | ASSERT(db->db.db_data); | |
66 | ASSERT(arc_released(db->db_buf)); | |
67 | ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); | |
68 | bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, | |
69 | sizeof (blkptr_t) * nblkptr); | |
70 | arc_buf_freeze(db->db_buf); | |
34dc7c2f BB |
71 | |
72 | /* set dbuf's parent pointers to new indirect buf */ | |
73 | for (i = 0; i < nblkptr; i++) { | |
6ebebace JG |
74 | dmu_buf_impl_t *child = |
75 | dbuf_find(dn->dn_objset, dn->dn_object, old_toplvl, i); | |
34dc7c2f BB |
76 | |
77 | if (child == NULL) | |
78 | continue; | |
572e2857 BB |
79 | #ifdef DEBUG |
80 | DB_DNODE_ENTER(child); | |
81 | ASSERT3P(DB_DNODE(child), ==, dn); | |
82 | DB_DNODE_EXIT(child); | |
83 | #endif /* DEBUG */ | |
34dc7c2f BB |
84 | if (child->db_parent && child->db_parent != dn->dn_dbuf) { |
85 | ASSERT(child->db_parent->db_level == db->db_level); | |
86 | ASSERT(child->db_blkptr != | |
87 | &dn->dn_phys->dn_blkptr[child->db_blkid]); | |
88 | mutex_exit(&child->db_mtx); | |
89 | continue; | |
90 | } | |
91 | ASSERT(child->db_parent == NULL || | |
92 | child->db_parent == dn->dn_dbuf); | |
93 | ||
94 | child->db_parent = db; | |
95 | dbuf_add_ref(db, child); | |
96 | if (db->db.db_data) | |
97 | child->db_blkptr = (blkptr_t *)db->db.db_data + i; | |
98 | else | |
99 | child->db_blkptr = NULL; | |
100 | dprintf_dbuf_bp(child, child->db_blkptr, | |
101 | "changed db_blkptr to new indirect %s", ""); | |
102 | ||
103 | mutex_exit(&child->db_mtx); | |
104 | } | |
105 | ||
106 | bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); | |
107 | ||
108 | dbuf_rele(db, FTAG); | |
109 | ||
110 | rw_exit(&dn->dn_struct_rwlock); | |
111 | } | |
112 | ||
b0bc7a84 | 113 | static void |
34dc7c2f BB |
114 | free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) |
115 | { | |
b128c09f | 116 | dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; |
34dc7c2f | 117 | uint64_t bytesfreed = 0; |
b0bc7a84 | 118 | int i; |
34dc7c2f | 119 | |
b128c09f | 120 | dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num); |
34dc7c2f BB |
121 | |
122 | for (i = 0; i < num; i++, bp++) { | |
b0bc7a84 MG |
123 | uint64_t lsize, lvl; |
124 | dmu_object_type_t type; | |
125 | ||
34dc7c2f BB |
126 | if (BP_IS_HOLE(bp)) |
127 | continue; | |
128 | ||
428870ff | 129 | bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE); |
34dc7c2f | 130 | ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); |
b0bc7a84 MG |
131 | |
132 | /* | |
133 | * Save some useful information on the holes being | |
134 | * punched, including logical size, type, and indirection | |
135 | * level. Retaining birth time enables detection of when | |
136 | * holes are punched for reducing the number of free | |
137 | * records transmitted during a zfs send. | |
138 | */ | |
139 | ||
140 | lsize = BP_GET_LSIZE(bp); | |
141 | type = BP_GET_TYPE(bp); | |
142 | lvl = BP_GET_LEVEL(bp); | |
143 | ||
34dc7c2f | 144 | bzero(bp, sizeof (blkptr_t)); |
b0bc7a84 MG |
145 | |
146 | if (spa_feature_is_active(dn->dn_objset->os_spa, | |
147 | SPA_FEATURE_HOLE_BIRTH)) { | |
148 | BP_SET_LSIZE(bp, lsize); | |
149 | BP_SET_TYPE(bp, type); | |
150 | BP_SET_LEVEL(bp, lvl); | |
151 | BP_SET_BIRTH(bp, dmu_tx_get_txg(tx), 0); | |
152 | } | |
34dc7c2f BB |
153 | } |
154 | dnode_diduse_space(dn, -bytesfreed); | |
155 | } | |
156 | ||
157 | #ifdef ZFS_DEBUG | |
158 | static void | |
159 | free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) | |
160 | { | |
161 | int off, num; | |
162 | int i, err, epbs; | |
163 | uint64_t txg = tx->tx_txg; | |
572e2857 | 164 | dnode_t *dn; |
34dc7c2f | 165 | |
572e2857 BB |
166 | DB_DNODE_ENTER(db); |
167 | dn = DB_DNODE(db); | |
168 | epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
34dc7c2f BB |
169 | off = start - (db->db_blkid * 1<<epbs); |
170 | num = end - start + 1; | |
171 | ||
172 | ASSERT3U(off, >=, 0); | |
173 | ASSERT3U(num, >=, 0); | |
174 | ASSERT3U(db->db_level, >, 0); | |
572e2857 | 175 | ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); |
34dc7c2f BB |
176 | ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); |
177 | ASSERT(db->db_blkptr != NULL); | |
178 | ||
179 | for (i = off; i < off+num; i++) { | |
180 | uint64_t *buf; | |
181 | dmu_buf_impl_t *child; | |
182 | dbuf_dirty_record_t *dr; | |
183 | int j; | |
184 | ||
185 | ASSERT(db->db_level == 1); | |
186 | ||
572e2857 BB |
187 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
188 | err = dbuf_hold_impl(dn, db->db_level-1, | |
fcff0f35 | 189 | (db->db_blkid << epbs) + i, TRUE, FALSE, FTAG, &child); |
572e2857 | 190 | rw_exit(&dn->dn_struct_rwlock); |
34dc7c2f BB |
191 | if (err == ENOENT) |
192 | continue; | |
193 | ASSERT(err == 0); | |
194 | ASSERT(child->db_level == 0); | |
195 | dr = child->db_last_dirty; | |
196 | while (dr && dr->dr_txg > txg) | |
197 | dr = dr->dr_next; | |
198 | ASSERT(dr == NULL || dr->dr_txg == txg); | |
199 | ||
200 | /* data_old better be zeroed */ | |
201 | if (dr) { | |
202 | buf = dr->dt.dl.dr_data->b_data; | |
203 | for (j = 0; j < child->db.db_size >> 3; j++) { | |
204 | if (buf[j] != 0) { | |
205 | panic("freed data not zero: " | |
206 | "child=%p i=%d off=%d num=%d\n", | |
b128c09f | 207 | (void *)child, i, off, num); |
34dc7c2f BB |
208 | } |
209 | } | |
210 | } | |
211 | ||
212 | /* | |
213 | * db_data better be zeroed unless it's dirty in a | |
214 | * future txg. | |
215 | */ | |
216 | mutex_enter(&child->db_mtx); | |
217 | buf = child->db.db_data; | |
218 | if (buf != NULL && child->db_state != DB_FILL && | |
219 | child->db_last_dirty == NULL) { | |
220 | for (j = 0; j < child->db.db_size >> 3; j++) { | |
221 | if (buf[j] != 0) { | |
222 | panic("freed data not zero: " | |
223 | "child=%p i=%d off=%d num=%d\n", | |
b128c09f | 224 | (void *)child, i, off, num); |
34dc7c2f BB |
225 | } |
226 | } | |
227 | } | |
228 | mutex_exit(&child->db_mtx); | |
229 | ||
230 | dbuf_rele(child, FTAG); | |
231 | } | |
572e2857 | 232 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
233 | } |
234 | #endif | |
235 | ||
b0bc7a84 MG |
236 | static void |
237 | free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, | |
34dc7c2f BB |
238 | dmu_tx_t *tx) |
239 | { | |
572e2857 | 240 | dnode_t *dn; |
34dc7c2f BB |
241 | blkptr_t *bp; |
242 | dmu_buf_impl_t *subdb; | |
721ed0ee GM |
243 | uint64_t start, end, dbstart, dbend; |
244 | unsigned int epbs, shift, i; | |
245 | uint64_t id; | |
b128c09f BB |
246 | |
247 | /* | |
248 | * There is a small possibility that this block will not be cached: | |
249 | * 1 - if level > 1 and there are no children with level <= 1 | |
b0bc7a84 MG |
250 | * 2 - if this block was evicted since we read it from |
251 | * dmu_tx_hold_free(). | |
b128c09f BB |
252 | */ |
253 | if (db->db_state != DB_CACHED) | |
254 | (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); | |
34dc7c2f | 255 | |
428870ff | 256 | dbuf_release_bp(db); |
b0bc7a84 | 257 | bp = db->db.db_data; |
34dc7c2f | 258 | |
572e2857 BB |
259 | DB_DNODE_ENTER(db); |
260 | dn = DB_DNODE(db); | |
261 | epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
721ed0ee | 262 | ASSERT3U(epbs, <, 31); |
34dc7c2f BB |
263 | shift = (db->db_level - 1) * epbs; |
264 | dbstart = db->db_blkid << epbs; | |
265 | start = blkid >> shift; | |
266 | if (dbstart < start) { | |
267 | bp += start - dbstart; | |
34dc7c2f BB |
268 | } else { |
269 | start = dbstart; | |
270 | } | |
271 | dbend = ((db->db_blkid + 1) << epbs) - 1; | |
272 | end = (blkid + nblks - 1) >> shift; | |
273 | if (dbend <= end) | |
274 | end = dbend; | |
b0bc7a84 | 275 | |
34dc7c2f BB |
276 | ASSERT3U(start, <=, end); |
277 | ||
278 | if (db->db_level == 1) { | |
279 | FREE_VERIFY(db, start, end, tx); | |
b0bc7a84 MG |
280 | free_blocks(dn, bp, end-start+1, tx); |
281 | } else { | |
721ed0ee | 282 | for (id = start; id <= end; id++, bp++) { |
b0bc7a84 MG |
283 | if (BP_IS_HOLE(bp)) |
284 | continue; | |
285 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
286 | VERIFY0(dbuf_hold_impl(dn, db->db_level - 1, | |
721ed0ee | 287 | id, TRUE, FALSE, FTAG, &subdb)); |
b0bc7a84 MG |
288 | rw_exit(&dn->dn_struct_rwlock); |
289 | ASSERT3P(bp, ==, subdb->db_blkptr); | |
290 | ||
291 | free_children(subdb, blkid, nblks, tx); | |
292 | dbuf_rele(subdb, FTAG); | |
293 | } | |
34dc7c2f BB |
294 | } |
295 | ||
b0bc7a84 | 296 | /* If this whole block is free, free ourself too. */ |
f4bae2ed | 297 | for (i = 0, bp = db->db.db_data; i < 1ULL << epbs; i++, bp++) { |
b0bc7a84 MG |
298 | if (!BP_IS_HOLE(bp)) |
299 | break; | |
300 | } | |
301 | if (i == 1 << epbs) { | |
721ed0ee GM |
302 | /* |
303 | * We only found holes. Grab the rwlock to prevent | |
304 | * anybody from reading the blocks we're about to | |
305 | * zero out. | |
306 | */ | |
307 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
b0bc7a84 | 308 | bzero(db->db.db_data, db->db.db_size); |
721ed0ee | 309 | rw_exit(&dn->dn_struct_rwlock); |
b0bc7a84 MG |
310 | free_blocks(dn, db->db_blkptr, 1, tx); |
311 | } else { | |
312 | /* | |
313 | * Partial block free; must be marked dirty so that it | |
314 | * will be written out. | |
315 | */ | |
316 | ASSERT(db->db_dirtycnt > 0); | |
34dc7c2f | 317 | } |
b0bc7a84 | 318 | |
572e2857 | 319 | DB_DNODE_EXIT(db); |
34dc7c2f | 320 | arc_buf_freeze(db->db_buf); |
34dc7c2f BB |
321 | } |
322 | ||
323 | /* | |
d3cc8b15 | 324 | * Traverse the indicated range of the provided file |
34dc7c2f BB |
325 | * and "free" all the blocks contained there. |
326 | */ | |
327 | static void | |
9bd274dd | 328 | dnode_sync_free_range_impl(dnode_t *dn, uint64_t blkid, uint64_t nblks, |
b0bc7a84 | 329 | dmu_tx_t *tx) |
34dc7c2f BB |
330 | { |
331 | blkptr_t *bp = dn->dn_phys->dn_blkptr; | |
34dc7c2f | 332 | int dnlevel = dn->dn_phys->dn_nlevels; |
b0bc7a84 | 333 | boolean_t trunc = B_FALSE; |
34dc7c2f BB |
334 | |
335 | if (blkid > dn->dn_phys->dn_maxblkid) | |
336 | return; | |
337 | ||
338 | ASSERT(dn->dn_phys->dn_maxblkid < UINT64_MAX); | |
b0bc7a84 | 339 | if (blkid + nblks > dn->dn_phys->dn_maxblkid) { |
34dc7c2f | 340 | nblks = dn->dn_phys->dn_maxblkid - blkid + 1; |
b0bc7a84 MG |
341 | trunc = B_TRUE; |
342 | } | |
34dc7c2f BB |
343 | |
344 | /* There are no indirect blocks in the object */ | |
345 | if (dnlevel == 1) { | |
346 | if (blkid >= dn->dn_phys->dn_nblkptr) { | |
347 | /* this range was never made persistent */ | |
348 | return; | |
349 | } | |
350 | ASSERT3U(blkid + nblks, <=, dn->dn_phys->dn_nblkptr); | |
b0bc7a84 MG |
351 | free_blocks(dn, bp + blkid, nblks, tx); |
352 | } else { | |
353 | int shift = (dnlevel - 1) * | |
354 | (dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT); | |
355 | int start = blkid >> shift; | |
356 | int end = (blkid + nblks - 1) >> shift; | |
357 | dmu_buf_impl_t *db; | |
358 | int i; | |
359 | ||
360 | ASSERT(start < dn->dn_phys->dn_nblkptr); | |
361 | bp += start; | |
362 | for (i = start; i <= end; i++, bp++) { | |
363 | if (BP_IS_HOLE(bp)) | |
364 | continue; | |
365 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
366 | VERIFY0(dbuf_hold_impl(dn, dnlevel - 1, i, | |
fcff0f35 | 367 | TRUE, FALSE, FTAG, &db)); |
b0bc7a84 MG |
368 | rw_exit(&dn->dn_struct_rwlock); |
369 | ||
370 | free_children(db, blkid, nblks, tx); | |
371 | dbuf_rele(db, FTAG); | |
34dc7c2f | 372 | } |
34dc7c2f BB |
373 | } |
374 | ||
34dc7c2f | 375 | if (trunc) { |
b0bc7a84 MG |
376 | ASSERTV(uint64_t off); |
377 | dn->dn_phys->dn_maxblkid = blkid == 0 ? 0 : blkid - 1; | |
378 | ||
379 | ASSERTV(off = (dn->dn_phys->dn_maxblkid + 1) * | |
1fde1e37 | 380 | (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT)); |
34dc7c2f BB |
381 | ASSERT(off < dn->dn_phys->dn_maxblkid || |
382 | dn->dn_phys->dn_maxblkid == 0 || | |
b128c09f | 383 | dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); |
34dc7c2f BB |
384 | } |
385 | } | |
386 | ||
9bd274dd MA |
387 | typedef struct dnode_sync_free_range_arg { |
388 | dnode_t *dsfra_dnode; | |
389 | dmu_tx_t *dsfra_tx; | |
390 | } dnode_sync_free_range_arg_t; | |
391 | ||
392 | static void | |
393 | dnode_sync_free_range(void *arg, uint64_t blkid, uint64_t nblks) | |
394 | { | |
395 | dnode_sync_free_range_arg_t *dsfra = arg; | |
396 | dnode_t *dn = dsfra->dsfra_dnode; | |
397 | ||
398 | mutex_exit(&dn->dn_mtx); | |
399 | dnode_sync_free_range_impl(dn, blkid, nblks, dsfra->dsfra_tx); | |
400 | mutex_enter(&dn->dn_mtx); | |
401 | } | |
402 | ||
34dc7c2f | 403 | /* |
d3cc8b15 | 404 | * Try to kick all the dnode's dbufs out of the cache... |
34dc7c2f BB |
405 | */ |
406 | void | |
407 | dnode_evict_dbufs(dnode_t *dn) | |
408 | { | |
0c66c32d JG |
409 | dmu_buf_impl_t *db_marker; |
410 | dmu_buf_impl_t *db, *db_next; | |
34dc7c2f | 411 | |
0c66c32d JG |
412 | db_marker = kmem_alloc(sizeof (dmu_buf_impl_t), KM_SLEEP); |
413 | ||
414 | mutex_enter(&dn->dn_dbufs_mtx); | |
415 | for (db = avl_first(&dn->dn_dbufs); db != NULL; db = db_next) { | |
34dc7c2f | 416 | |
572e2857 | 417 | #ifdef DEBUG |
0c66c32d JG |
418 | DB_DNODE_ENTER(db); |
419 | ASSERT3P(DB_DNODE(db), ==, dn); | |
420 | DB_DNODE_EXIT(db); | |
572e2857 | 421 | #endif /* DEBUG */ |
34dc7c2f | 422 | |
0c66c32d JG |
423 | mutex_enter(&db->db_mtx); |
424 | if (db->db_state != DB_EVICTING && | |
b884768e | 425 | zfs_refcount_is_zero(&db->db_holds)) { |
0c66c32d JG |
426 | db_marker->db_level = db->db_level; |
427 | db_marker->db_blkid = db->db_blkid; | |
428 | db_marker->db_state = DB_SEARCH; | |
429 | avl_insert_here(&dn->dn_dbufs, db_marker, db, | |
430 | AVL_BEFORE); | |
431 | ||
01937958 MA |
432 | /* |
433 | * We need to use the "marker" dbuf rather than | |
434 | * simply getting the next dbuf, because | |
435 | * dbuf_destroy() may actually remove multiple dbufs. | |
436 | * It can call itself recursively on the parent dbuf, | |
437 | * which may also be removed from dn_dbufs. The code | |
438 | * flow would look like: | |
439 | * | |
440 | * dbuf_destroy(): | |
441 | * dnode_rele_and_unlock(parent_dbuf, evicting=TRUE): | |
442 | * if (!cacheable || pending_evict) | |
443 | * dbuf_destroy() | |
444 | */ | |
d3c2ae1c | 445 | dbuf_destroy(db); |
0c66c32d JG |
446 | |
447 | db_next = AVL_NEXT(&dn->dn_dbufs, db_marker); | |
448 | avl_remove(&dn->dn_dbufs, db_marker); | |
449 | } else { | |
bc4501f7 | 450 | db->db_pending_evict = TRUE; |
0c66c32d JG |
451 | mutex_exit(&db->db_mtx); |
452 | db_next = AVL_NEXT(&dn->dn_dbufs, db); | |
34dc7c2f | 453 | } |
0c66c32d JG |
454 | } |
455 | mutex_exit(&dn->dn_dbufs_mtx); | |
34dc7c2f | 456 | |
0c66c32d | 457 | kmem_free(db_marker, sizeof (dmu_buf_impl_t)); |
754c6663 | 458 | |
4c7b7eed JG |
459 | dnode_evict_bonus(dn); |
460 | } | |
461 | ||
462 | void | |
463 | dnode_evict_bonus(dnode_t *dn) | |
464 | { | |
34dc7c2f | 465 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); |
bc4501f7 | 466 | if (dn->dn_bonus != NULL) { |
b884768e | 467 | if (zfs_refcount_is_zero(&dn->dn_bonus->db_holds)) { |
bc4501f7 | 468 | mutex_enter(&dn->dn_bonus->db_mtx); |
d3c2ae1c | 469 | dbuf_destroy(dn->dn_bonus); |
bc4501f7 JG |
470 | dn->dn_bonus = NULL; |
471 | } else { | |
472 | dn->dn_bonus->db_pending_evict = TRUE; | |
473 | } | |
34dc7c2f BB |
474 | } |
475 | rw_exit(&dn->dn_struct_rwlock); | |
476 | } | |
477 | ||
478 | static void | |
479 | dnode_undirty_dbufs(list_t *list) | |
480 | { | |
481 | dbuf_dirty_record_t *dr; | |
482 | ||
c65aa5b2 | 483 | while ((dr = list_head(list))) { |
34dc7c2f BB |
484 | dmu_buf_impl_t *db = dr->dr_dbuf; |
485 | uint64_t txg = dr->dr_txg; | |
486 | ||
428870ff BB |
487 | if (db->db_level != 0) |
488 | dnode_undirty_dbufs(&dr->dt.di.dr_children); | |
489 | ||
34dc7c2f BB |
490 | mutex_enter(&db->db_mtx); |
491 | /* XXX - use dbuf_undirty()? */ | |
492 | list_remove(list, dr); | |
493 | ASSERT(db->db_last_dirty == dr); | |
494 | db->db_last_dirty = NULL; | |
495 | db->db_dirtycnt -= 1; | |
496 | if (db->db_level == 0) { | |
428870ff | 497 | ASSERT(db->db_blkid == DMU_BONUS_BLKID || |
34dc7c2f BB |
498 | dr->dt.dl.dr_data == db->db_buf); |
499 | dbuf_unoverride(dr); | |
58c4aa00 JL |
500 | } else { |
501 | mutex_destroy(&dr->dt.di.dr_mtx); | |
502 | list_destroy(&dr->dt.di.dr_children); | |
34dc7c2f BB |
503 | } |
504 | kmem_free(dr, sizeof (dbuf_dirty_record_t)); | |
01937958 | 505 | dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg, B_FALSE); |
34dc7c2f BB |
506 | } |
507 | } | |
508 | ||
509 | static void | |
510 | dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) | |
511 | { | |
512 | int txgoff = tx->tx_txg & TXG_MASK; | |
513 | ||
514 | ASSERT(dmu_tx_is_syncing(tx)); | |
515 | ||
b128c09f BB |
516 | /* |
517 | * Our contents should have been freed in dnode_sync() by the | |
518 | * free range record inserted by the caller of dnode_free(). | |
519 | */ | |
c99c9001 | 520 | ASSERT0(DN_USED_BYTES(dn->dn_phys)); |
b128c09f BB |
521 | ASSERT(BP_IS_HOLE(dn->dn_phys->dn_blkptr)); |
522 | ||
34dc7c2f BB |
523 | dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); |
524 | dnode_evict_dbufs(dn); | |
34dc7c2f BB |
525 | |
526 | /* | |
527 | * XXX - It would be nice to assert this, but we may still | |
528 | * have residual holds from async evictions from the arc... | |
529 | * | |
530 | * zfs_obj_to_path() also depends on this being | |
531 | * commented out. | |
532 | * | |
b884768e | 533 | * ASSERT3U(zfs_refcount_count(&dn->dn_holds), ==, 1); |
34dc7c2f BB |
534 | */ |
535 | ||
536 | /* Undirty next bits */ | |
537 | dn->dn_next_nlevels[txgoff] = 0; | |
538 | dn->dn_next_indblkshift[txgoff] = 0; | |
539 | dn->dn_next_blksz[txgoff] = 0; | |
540 | ||
34dc7c2f BB |
541 | /* ASSERT(blkptrs are zero); */ |
542 | ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); | |
543 | ASSERT(dn->dn_type != DMU_OT_NONE); | |
544 | ||
545 | ASSERT(dn->dn_free_txg > 0); | |
546 | if (dn->dn_allocated_txg != dn->dn_free_txg) | |
b0bc7a84 | 547 | dmu_buf_will_dirty(&dn->dn_dbuf->db, tx); |
50c957f7 | 548 | bzero(dn->dn_phys, sizeof (dnode_phys_t) * dn->dn_num_slots); |
d2c8103a | 549 | dnode_free_interior_slots(dn); |
34dc7c2f BB |
550 | |
551 | mutex_enter(&dn->dn_mtx); | |
552 | dn->dn_type = DMU_OT_NONE; | |
553 | dn->dn_maxblkid = 0; | |
554 | dn->dn_allocated_txg = 0; | |
555 | dn->dn_free_txg = 0; | |
428870ff | 556 | dn->dn_have_spill = B_FALSE; |
d2c8103a | 557 | dn->dn_num_slots = 1; |
34dc7c2f BB |
558 | mutex_exit(&dn->dn_mtx); |
559 | ||
560 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); | |
561 | ||
562 | dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); | |
563 | /* | |
564 | * Now that we've released our hold, the dnode may | |
4e33ba4c | 565 | * be evicted, so we mustn't access it. |
34dc7c2f BB |
566 | */ |
567 | } | |
568 | ||
569 | /* | |
570 | * Write out the dnode's dirty buffers. | |
34dc7c2f BB |
571 | */ |
572 | void | |
573 | dnode_sync(dnode_t *dn, dmu_tx_t *tx) | |
574 | { | |
34dc7c2f BB |
575 | dnode_phys_t *dnp = dn->dn_phys; |
576 | int txgoff = tx->tx_txg & TXG_MASK; | |
577 | list_t *list = &dn->dn_dirty_records[txgoff]; | |
428870ff | 578 | boolean_t kill_spill = B_FALSE; |
b0bc7a84 | 579 | boolean_t freeing_dnode; |
1fde1e37 | 580 | ASSERTV(static const dnode_phys_t zerodn = { 0 }); |
34dc7c2f BB |
581 | |
582 | ASSERT(dmu_tx_is_syncing(tx)); | |
583 | ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg); | |
9babb374 | 584 | ASSERT(dnp->dn_type != DMU_OT_NONE || |
50c957f7 | 585 | bcmp(dnp, &zerodn, DNODE_MIN_SIZE) == 0); |
34dc7c2f BB |
586 | DNODE_VERIFY(dn); |
587 | ||
588 | ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf)); | |
589 | ||
9babb374 BB |
590 | if (dmu_objset_userused_enabled(dn->dn_objset) && |
591 | !DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { | |
428870ff BB |
592 | mutex_enter(&dn->dn_mtx); |
593 | dn->dn_oldused = DN_USED_BYTES(dn->dn_phys); | |
594 | dn->dn_oldflags = dn->dn_phys->dn_flags; | |
9babb374 | 595 | dn->dn_phys->dn_flags |= DNODE_FLAG_USERUSED_ACCOUNTED; |
1de321e6 JX |
596 | if (dmu_objset_userobjused_enabled(dn->dn_objset)) |
597 | dn->dn_phys->dn_flags |= | |
598 | DNODE_FLAG_USEROBJUSED_ACCOUNTED; | |
428870ff BB |
599 | mutex_exit(&dn->dn_mtx); |
600 | dmu_objset_userquota_get_ids(dn, B_FALSE, tx); | |
9babb374 BB |
601 | } else { |
602 | /* Once we account for it, we should always account for it. */ | |
603 | ASSERT(!(dn->dn_phys->dn_flags & | |
604 | DNODE_FLAG_USERUSED_ACCOUNTED)); | |
1de321e6 JX |
605 | ASSERT(!(dn->dn_phys->dn_flags & |
606 | DNODE_FLAG_USEROBJUSED_ACCOUNTED)); | |
9babb374 BB |
607 | } |
608 | ||
34dc7c2f BB |
609 | mutex_enter(&dn->dn_mtx); |
610 | if (dn->dn_allocated_txg == tx->tx_txg) { | |
611 | /* The dnode is newly allocated or reallocated */ | |
612 | if (dnp->dn_type == DMU_OT_NONE) { | |
613 | /* this is a first alloc, not a realloc */ | |
34dc7c2f | 614 | dnp->dn_nlevels = 1; |
d164b209 | 615 | dnp->dn_nblkptr = dn->dn_nblkptr; |
34dc7c2f BB |
616 | } |
617 | ||
34dc7c2f BB |
618 | dnp->dn_type = dn->dn_type; |
619 | dnp->dn_bonustype = dn->dn_bonustype; | |
620 | dnp->dn_bonuslen = dn->dn_bonuslen; | |
34dc7c2f | 621 | } |
50c957f7 NB |
622 | |
623 | dnp->dn_extra_slots = dn->dn_num_slots - 1; | |
624 | ||
34dc7c2f BB |
625 | ASSERT(dnp->dn_nlevels > 1 || |
626 | BP_IS_HOLE(&dnp->dn_blkptr[0]) || | |
9b67f605 | 627 | BP_IS_EMBEDDED(&dnp->dn_blkptr[0]) || |
34dc7c2f BB |
628 | BP_GET_LSIZE(&dnp->dn_blkptr[0]) == |
629 | dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
9b67f605 MA |
630 | ASSERT(dnp->dn_nlevels < 2 || |
631 | BP_IS_HOLE(&dnp->dn_blkptr[0]) || | |
632 | BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 1 << dnp->dn_indblkshift); | |
34dc7c2f | 633 | |
fa86b5db MA |
634 | if (dn->dn_next_type[txgoff] != 0) { |
635 | dnp->dn_type = dn->dn_type; | |
636 | dn->dn_next_type[txgoff] = 0; | |
637 | } | |
638 | ||
639 | if (dn->dn_next_blksz[txgoff] != 0) { | |
34dc7c2f BB |
640 | ASSERT(P2PHASE(dn->dn_next_blksz[txgoff], |
641 | SPA_MINBLOCKSIZE) == 0); | |
642 | ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[0]) || | |
b128c09f | 643 | dn->dn_maxblkid == 0 || list_head(list) != NULL || |
34dc7c2f | 644 | dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT == |
9bd274dd MA |
645 | dnp->dn_datablkszsec || |
646 | range_tree_space(dn->dn_free_ranges[txgoff]) != 0); | |
34dc7c2f BB |
647 | dnp->dn_datablkszsec = |
648 | dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT; | |
649 | dn->dn_next_blksz[txgoff] = 0; | |
650 | } | |
651 | ||
fa86b5db | 652 | if (dn->dn_next_bonuslen[txgoff] != 0) { |
34dc7c2f BB |
653 | if (dn->dn_next_bonuslen[txgoff] == DN_ZERO_BONUSLEN) |
654 | dnp->dn_bonuslen = 0; | |
655 | else | |
656 | dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff]; | |
50c957f7 NB |
657 | ASSERT(dnp->dn_bonuslen <= |
658 | DN_SLOTS_TO_BONUSLEN(dnp->dn_extra_slots + 1)); | |
34dc7c2f BB |
659 | dn->dn_next_bonuslen[txgoff] = 0; |
660 | } | |
661 | ||
fa86b5db | 662 | if (dn->dn_next_bonustype[txgoff] != 0) { |
9ae529ec | 663 | ASSERT(DMU_OT_IS_VALID(dn->dn_next_bonustype[txgoff])); |
428870ff BB |
664 | dnp->dn_bonustype = dn->dn_next_bonustype[txgoff]; |
665 | dn->dn_next_bonustype[txgoff] = 0; | |
666 | } | |
667 | ||
b0bc7a84 MG |
668 | freeing_dnode = dn->dn_free_txg > 0 && dn->dn_free_txg <= tx->tx_txg; |
669 | ||
428870ff | 670 | /* |
08dc1b2d MA |
671 | * Remove the spill block if we have been explicitly asked to |
672 | * remove it, or if the object is being removed. | |
428870ff | 673 | */ |
08dc1b2d MA |
674 | if (dn->dn_rm_spillblk[txgoff] || freeing_dnode) { |
675 | if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) | |
428870ff BB |
676 | kill_spill = B_TRUE; |
677 | dn->dn_rm_spillblk[txgoff] = 0; | |
678 | } | |
679 | ||
fa86b5db | 680 | if (dn->dn_next_indblkshift[txgoff] != 0) { |
34dc7c2f BB |
681 | ASSERT(dnp->dn_nlevels == 1); |
682 | dnp->dn_indblkshift = dn->dn_next_indblkshift[txgoff]; | |
683 | dn->dn_next_indblkshift[txgoff] = 0; | |
684 | } | |
685 | ||
686 | /* | |
687 | * Just take the live (open-context) values for checksum and compress. | |
688 | * Strictly speaking it's a future leak, but nothing bad happens if we | |
689 | * start using the new checksum or compress algorithm a little early. | |
690 | */ | |
691 | dnp->dn_checksum = dn->dn_checksum; | |
692 | dnp->dn_compress = dn->dn_compress; | |
693 | ||
694 | mutex_exit(&dn->dn_mtx); | |
695 | ||
428870ff | 696 | if (kill_spill) { |
50c957f7 | 697 | free_blocks(dn, DN_SPILL_BLKPTR(dn->dn_phys), 1, tx); |
428870ff BB |
698 | mutex_enter(&dn->dn_mtx); |
699 | dnp->dn_flags &= ~DNODE_FLAG_SPILL_BLKPTR; | |
700 | mutex_exit(&dn->dn_mtx); | |
701 | } | |
702 | ||
34dc7c2f | 703 | /* process all the "freed" ranges in the file */ |
9bd274dd MA |
704 | if (dn->dn_free_ranges[txgoff] != NULL) { |
705 | dnode_sync_free_range_arg_t dsfra; | |
706 | dsfra.dsfra_dnode = dn; | |
707 | dsfra.dsfra_tx = tx; | |
b128c09f | 708 | mutex_enter(&dn->dn_mtx); |
9bd274dd MA |
709 | range_tree_vacate(dn->dn_free_ranges[txgoff], |
710 | dnode_sync_free_range, &dsfra); | |
711 | range_tree_destroy(dn->dn_free_ranges[txgoff]); | |
712 | dn->dn_free_ranges[txgoff] = NULL; | |
b128c09f | 713 | mutex_exit(&dn->dn_mtx); |
34dc7c2f | 714 | } |
34dc7c2f | 715 | |
b0bc7a84 | 716 | if (freeing_dnode) { |
68cbd56e | 717 | dn->dn_objset->os_freed_dnodes++; |
34dc7c2f BB |
718 | dnode_sync_free(dn, tx); |
719 | return; | |
720 | } | |
721 | ||
50c957f7 NB |
722 | if (dn->dn_num_slots > DNODE_MIN_SLOTS) { |
723 | dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; | |
724 | mutex_enter(&ds->ds_lock); | |
725 | ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_DNODE] = | |
726 | B_TRUE; | |
727 | mutex_exit(&ds->ds_lock); | |
728 | } | |
729 | ||
4f68d787 GW |
730 | if (dn->dn_next_nlevels[txgoff]) { |
731 | dnode_increase_indirection(dn, tx); | |
732 | dn->dn_next_nlevels[txgoff] = 0; | |
733 | } | |
734 | ||
d164b209 BB |
735 | if (dn->dn_next_nblkptr[txgoff]) { |
736 | /* this should only happen on a realloc */ | |
737 | ASSERT(dn->dn_allocated_txg == tx->tx_txg); | |
738 | if (dn->dn_next_nblkptr[txgoff] > dnp->dn_nblkptr) { | |
739 | /* zero the new blkptrs we are gaining */ | |
740 | bzero(dnp->dn_blkptr + dnp->dn_nblkptr, | |
741 | sizeof (blkptr_t) * | |
742 | (dn->dn_next_nblkptr[txgoff] - dnp->dn_nblkptr)); | |
743 | #ifdef ZFS_DEBUG | |
744 | } else { | |
745 | int i; | |
746 | ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr); | |
747 | /* the blkptrs we are losing better be unallocated */ | |
37f000c5 NB |
748 | for (i = 0; i < dnp->dn_nblkptr; i++) { |
749 | if (i >= dn->dn_next_nblkptr[txgoff]) | |
750 | ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i])); | |
751 | } | |
d164b209 BB |
752 | #endif |
753 | } | |
754 | mutex_enter(&dn->dn_mtx); | |
755 | dnp->dn_nblkptr = dn->dn_next_nblkptr[txgoff]; | |
756 | dn->dn_next_nblkptr[txgoff] = 0; | |
757 | mutex_exit(&dn->dn_mtx); | |
758 | } | |
759 | ||
4bda3bd0 | 760 | dbuf_sync_list(list, dn->dn_phys->dn_nlevels - 1, tx); |
34dc7c2f | 761 | |
9babb374 | 762 | if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { |
34dc7c2f BB |
763 | ASSERT3P(list_head(list), ==, NULL); |
764 | dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); | |
765 | } | |
766 | ||
767 | /* | |
768 | * Although we have dropped our reference to the dnode, it | |
769 | * can't be evicted until its written, and we haven't yet | |
770 | * initiated the IO for the dnode's dbuf. | |
771 | */ | |
772 | } |