Deleted Added
sdiff udiff text old ( 263397 ) new ( 268650 )
full compact
bptree.c (263397) bptree.c (268650)
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

--- 88 unchanged lines hidden (view full) ---

97 ASSERT0(bt->bt_bytes);
98 ASSERT0(bt->bt_comp);
99 ASSERT0(bt->bt_uncomp);
100 dmu_buf_rele(db, FTAG);
101
102 return (dmu_object_free(os, obj, tx));
103}
104
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

--- 88 unchanged lines hidden (view full) ---

97 ASSERT0(bt->bt_bytes);
98 ASSERT0(bt->bt_comp);
99 ASSERT0(bt->bt_uncomp);
100 dmu_buf_rele(db, FTAG);
101
102 return (dmu_object_free(os, obj, tx));
103}
104
105boolean_t
106bptree_is_empty(objset_t *os, uint64_t obj)
107{
108 dmu_buf_t *db;
109 bptree_phys_t *bt;
110 boolean_t rv;
111
112 VERIFY0(dmu_bonus_hold(os, obj, FTAG, &db));
113 bt = db->db_data;
114 rv = (bt->bt_begin == bt->bt_end);
115 dmu_buf_rele(db, FTAG);
116 return (rv);
117}
118
105void
106bptree_add(objset_t *os, uint64_t obj, blkptr_t *bp, uint64_t birth_txg,
107 uint64_t bytes, uint64_t comp, uint64_t uncomp, dmu_tx_t *tx)
108{
109 dmu_buf_t *db;
110 bptree_phys_t *bt;
119void
120bptree_add(objset_t *os, uint64_t obj, blkptr_t *bp, uint64_t birth_txg,
121 uint64_t bytes, uint64_t comp, uint64_t uncomp, dmu_tx_t *tx)
122{
123 dmu_buf_t *db;
124 bptree_phys_t *bt;
111 bptree_entry_phys_t bte;
125 bptree_entry_phys_t bte = { 0 };
112
113 /*
114 * bptree objects are in the pool mos, therefore they can only be
115 * modified in syncing context. Furthermore, this is only modified
116 * by the sync thread, so no locking is necessary.
117 */
118 ASSERT(dmu_tx_is_syncing(tx));
119
120 VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
121 bt = db->db_data;
122
123 bte.be_birth_txg = birth_txg;
124 bte.be_bp = *bp;
126
127 /*
128 * bptree objects are in the pool mos, therefore they can only be
129 * modified in syncing context. Furthermore, this is only modified
130 * by the sync thread, so no locking is necessary.
131 */
132 ASSERT(dmu_tx_is_syncing(tx));
133
134 VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
135 bt = db->db_data;
136
137 bte.be_birth_txg = birth_txg;
138 bte.be_bp = *bp;
125 bzero(&bte.be_zb, sizeof (bte.be_zb));
126 dmu_write(os, obj, bt->bt_end * sizeof (bte), sizeof (bte), &bte, tx);
127
128 dmu_buf_will_dirty(db, tx);
129 bt->bt_end++;
130 bt->bt_bytes += bytes;
131 bt->bt_comp += comp;
132 bt->bt_uncomp += uncomp;
133 dmu_buf_rele(db, FTAG);

--- 14 unchanged lines hidden (view full) ---

148 if (err == 0 && ba->ba_free) {
149 ba->ba_phys->bt_bytes -= bp_get_dsize_sync(spa, bp);
150 ba->ba_phys->bt_comp -= BP_GET_PSIZE(bp);
151 ba->ba_phys->bt_uncomp -= BP_GET_UCSIZE(bp);
152 }
153 return (err);
154}
155
139 dmu_write(os, obj, bt->bt_end * sizeof (bte), sizeof (bte), &bte, tx);
140
141 dmu_buf_will_dirty(db, tx);
142 bt->bt_end++;
143 bt->bt_bytes += bytes;
144 bt->bt_comp += comp;
145 bt->bt_uncomp += uncomp;
146 dmu_buf_rele(db, FTAG);

--- 14 unchanged lines hidden (view full) ---

161 if (err == 0 && ba->ba_free) {
162 ba->ba_phys->bt_bytes -= bp_get_dsize_sync(spa, bp);
163 ba->ba_phys->bt_comp -= BP_GET_PSIZE(bp);
164 ba->ba_phys->bt_uncomp -= BP_GET_UCSIZE(bp);
165 }
166 return (err);
167}
168
169/*
170 * If "free" is set:
171 * - It is assumed that "func" will be freeing the block pointers.
172 * - If "func" returns nonzero, the bookmark will be remembered and
173 * iteration will be restarted from this point on next invocation.
174 * - If an i/o error is encountered (e.g. "func" returns EIO or ECKSUM),
175 * bptree_iterate will remember the bookmark, continue traversing
176 * any additional entries, and return 0.
177 *
178 * If "free" is not set, traversal will stop and return an error if
179 * an i/o error is encountered.
180 *
181 * In either case, if zfs_free_leak_on_eio is set, i/o errors will be
182 * ignored and traversal will continue (i.e. TRAVERSE_HARD will be passed to
183 * traverse_dataset_destroyed()).
184 */
156int
157bptree_iterate(objset_t *os, uint64_t obj, boolean_t free, bptree_itor_t func,
158 void *arg, dmu_tx_t *tx)
159{
185int
186bptree_iterate(objset_t *os, uint64_t obj, boolean_t free, bptree_itor_t func,
187 void *arg, dmu_tx_t *tx)
188{
189 boolean_t ioerr = B_FALSE;
160 int err;
161 uint64_t i;
162 dmu_buf_t *db;
163 struct bptree_args ba;
164
165 ASSERT(!free || dmu_tx_is_syncing(tx));
166
167 err = dmu_bonus_hold(os, obj, FTAG, &db);

--- 9 unchanged lines hidden (view full) ---

177 ba.ba_arg = arg;
178 ba.ba_tx = tx;
179
180 err = 0;
181 for (i = ba.ba_phys->bt_begin; i < ba.ba_phys->bt_end; i++) {
182 bptree_entry_phys_t bte;
183 int flags = TRAVERSE_PREFETCH_METADATA | TRAVERSE_POST;
184
190 int err;
191 uint64_t i;
192 dmu_buf_t *db;
193 struct bptree_args ba;
194
195 ASSERT(!free || dmu_tx_is_syncing(tx));
196
197 err = dmu_bonus_hold(os, obj, FTAG, &db);

--- 9 unchanged lines hidden (view full) ---

207 ba.ba_arg = arg;
208 ba.ba_tx = tx;
209
210 err = 0;
211 for (i = ba.ba_phys->bt_begin; i < ba.ba_phys->bt_end; i++) {
212 bptree_entry_phys_t bte;
213 int flags = TRAVERSE_PREFETCH_METADATA | TRAVERSE_POST;
214
185 ASSERT(!free || i == ba.ba_phys->bt_begin);
186
187 err = dmu_read(os, obj, i * sizeof (bte), sizeof (bte),
188 &bte, DMU_READ_NO_PREFETCH);
189 if (err != 0)
190 break;
191
215 err = dmu_read(os, obj, i * sizeof (bte), sizeof (bte),
216 &bte, DMU_READ_NO_PREFETCH);
217 if (err != 0)
218 break;
219
192 if (zfs_recover)
220 if (zfs_free_leak_on_eio)
193 flags |= TRAVERSE_HARD;
221 flags |= TRAVERSE_HARD;
222 zfs_dbgmsg("bptree index %d: traversing from min_txg=%lld "
223 "bookmark %lld/%lld/%lld/%lld",
224 i, (longlong_t)bte.be_birth_txg,
225 (longlong_t)bte.be_zb.zb_objset,
226 (longlong_t)bte.be_zb.zb_object,
227 (longlong_t)bte.be_zb.zb_level,
228 (longlong_t)bte.be_zb.zb_blkid);
194 err = traverse_dataset_destroyed(os->os_spa, &bte.be_bp,
195 bte.be_birth_txg, &bte.be_zb, flags,
196 bptree_visit_cb, &ba);
197 if (free) {
229 err = traverse_dataset_destroyed(os->os_spa, &bte.be_bp,
230 bte.be_birth_txg, &bte.be_zb, flags,
231 bptree_visit_cb, &ba);
232 if (free) {
198 if (err == ERESTART) {
233 /*
234 * The callback has freed the visited block pointers.
235 * Record our traversal progress on disk, either by
236 * updating this record's bookmark, or by logically
237 * removing this record by advancing bt_begin.
238 */
239 if (err != 0) {
199 /* save bookmark for future resume */
200 ASSERT3U(bte.be_zb.zb_objset, ==,
201 ZB_DESTROYED_OBJSET);
202 ASSERT0(bte.be_zb.zb_level);
203 dmu_write(os, obj, i * sizeof (bte),
204 sizeof (bte), &bte, tx);
240 /* save bookmark for future resume */
241 ASSERT3U(bte.be_zb.zb_objset, ==,
242 ZB_DESTROYED_OBJSET);
243 ASSERT0(bte.be_zb.zb_level);
244 dmu_write(os, obj, i * sizeof (bte),
245 sizeof (bte), &bte, tx);
205 break;
206 }
207 if (err != 0) {
246 if (err == EIO || err == ECKSUM ||
247 err == ENXIO) {
248 /*
249 * Skip the rest of this tree and
250 * continue on to the next entry.
251 */
252 err = 0;
253 ioerr = B_TRUE;
254 } else {
255 break;
256 }
257 } else if (ioerr) {
208 /*
258 /*
209 * We can not properly handle an i/o
210 * error, because the traversal code
211 * does not know how to resume from an
212 * arbitrary bookmark.
259 * This entry is finished, but there were
260 * i/o errors on previous entries, so we
261 * can't adjust bt_begin. Set this entry's
262 * be_birth_txg such that it will be
263 * treated as a no-op in future traversals.
213 */
264 */
214 zfs_panic_recover("error %u from "
215 "traverse_dataset_destroyed()", err);
265 bte.be_birth_txg = UINT64_MAX;
266 dmu_write(os, obj, i * sizeof (bte),
267 sizeof (bte), &bte, tx);
216 }
217
268 }
269
218 ba.ba_phys->bt_begin++;
219 (void) dmu_free_range(os, obj,
220 i * sizeof (bte), sizeof (bte), tx);
270 if (!ioerr) {
271 ba.ba_phys->bt_begin++;
272 (void) dmu_free_range(os, obj,
273 i * sizeof (bte), sizeof (bte), tx);
274 }
275 } else if (err != 0) {
276 break;
221 }
222 }
223
277 }
278 }
279
224 ASSERT(!free || err != 0 || ba.ba_phys->bt_begin == ba.ba_phys->bt_end);
280 ASSERT(!free || err != 0 || ioerr ||
281 ba.ba_phys->bt_begin == ba.ba_phys->bt_end);
225
226 /* if all blocks are free there should be no used space */
227 if (ba.ba_phys->bt_begin == ba.ba_phys->bt_end) {
282
283 /* if all blocks are free there should be no used space */
284 if (ba.ba_phys->bt_begin == ba.ba_phys->bt_end) {
285 if (zfs_free_leak_on_eio) {
286 ba.ba_phys->bt_bytes = 0;
287 ba.ba_phys->bt_comp = 0;
288 ba.ba_phys->bt_uncomp = 0;
289 }
290
228 ASSERT0(ba.ba_phys->bt_bytes);
229 ASSERT0(ba.ba_phys->bt_comp);
230 ASSERT0(ba.ba_phys->bt_uncomp);
231 }
232
233 dmu_buf_rele(db, FTAG);
234
235 return (err);
236}
291 ASSERT0(ba.ba_phys->bt_bytes);
292 ASSERT0(ba.ba_phys->bt_comp);
293 ASSERT0(ba.ba_phys->bt_uncomp);
294 }
295
296 dmu_buf_rele(db, FTAG);
297
298 return (err);
299}