]>
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 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
22cd4a46 | 23 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. |
295304be | 24 | * Copyright (c) 2013 by Delphix. All rights reserved. |
22cd4a46 | 25 | */ |
34dc7c2f | 26 | |
34dc7c2f BB |
27 | #include <sys/dmu.h> |
28 | #include <sys/dmu_impl.h> | |
29 | #include <sys/dbuf.h> | |
30 | #include <sys/dmu_tx.h> | |
31 | #include <sys/dmu_objset.h> | |
32 | #include <sys/dsl_dataset.h> /* for dsl_dataset_block_freeable() */ | |
33 | #include <sys/dsl_dir.h> /* for dsl_dir_tempreserve_*() */ | |
34 | #include <sys/dsl_pool.h> | |
35 | #include <sys/zap_impl.h> /* for fzap_default_block_shift */ | |
36 | #include <sys/spa.h> | |
428870ff BB |
37 | #include <sys/sa.h> |
38 | #include <sys/sa_impl.h> | |
34dc7c2f | 39 | #include <sys/zfs_context.h> |
428870ff | 40 | #include <sys/varargs.h> |
34dc7c2f BB |
41 | |
42 | typedef void (*dmu_tx_hold_func_t)(dmu_tx_t *tx, struct dnode *dn, | |
43 | uint64_t arg1, uint64_t arg2); | |
44 | ||
570827e1 BB |
45 | dmu_tx_stats_t dmu_tx_stats = { |
46 | { "dmu_tx_assigned", KSTAT_DATA_UINT64 }, | |
47 | { "dmu_tx_delay", KSTAT_DATA_UINT64 }, | |
48 | { "dmu_tx_error", KSTAT_DATA_UINT64 }, | |
49 | { "dmu_tx_suspended", KSTAT_DATA_UINT64 }, | |
50 | { "dmu_tx_group", KSTAT_DATA_UINT64 }, | |
51 | { "dmu_tx_how", KSTAT_DATA_UINT64 }, | |
52 | { "dmu_tx_memory_reserve", KSTAT_DATA_UINT64 }, | |
53 | { "dmu_tx_memory_reclaim", KSTAT_DATA_UINT64 }, | |
54 | { "dmu_tx_memory_inflight", KSTAT_DATA_UINT64 }, | |
55 | { "dmu_tx_dirty_throttle", KSTAT_DATA_UINT64 }, | |
56 | { "dmu_tx_write_limit", KSTAT_DATA_UINT64 }, | |
57 | { "dmu_tx_quota", KSTAT_DATA_UINT64 }, | |
58 | }; | |
59 | ||
60 | static kstat_t *dmu_tx_ksp; | |
34dc7c2f BB |
61 | |
62 | dmu_tx_t * | |
63 | dmu_tx_create_dd(dsl_dir_t *dd) | |
64 | { | |
b8d06fca | 65 | dmu_tx_t *tx = kmem_zalloc(sizeof (dmu_tx_t), KM_PUSHPAGE); |
34dc7c2f BB |
66 | tx->tx_dir = dd; |
67 | if (dd) | |
68 | tx->tx_pool = dd->dd_pool; | |
69 | list_create(&tx->tx_holds, sizeof (dmu_tx_hold_t), | |
70 | offsetof(dmu_tx_hold_t, txh_node)); | |
428870ff BB |
71 | list_create(&tx->tx_callbacks, sizeof (dmu_tx_callback_t), |
72 | offsetof(dmu_tx_callback_t, dcb_node)); | |
1c5de20a | 73 | #ifdef DEBUG_DMU_TX |
34dc7c2f BB |
74 | refcount_create(&tx->tx_space_written); |
75 | refcount_create(&tx->tx_space_freed); | |
76 | #endif | |
77 | return (tx); | |
78 | } | |
79 | ||
80 | dmu_tx_t * | |
81 | dmu_tx_create(objset_t *os) | |
82 | { | |
428870ff | 83 | dmu_tx_t *tx = dmu_tx_create_dd(os->os_dsl_dataset->ds_dir); |
34dc7c2f | 84 | tx->tx_objset = os; |
428870ff | 85 | tx->tx_lastsnap_txg = dsl_dataset_prev_snap_txg(os->os_dsl_dataset); |
34dc7c2f BB |
86 | return (tx); |
87 | } | |
88 | ||
89 | dmu_tx_t * | |
90 | dmu_tx_create_assigned(struct dsl_pool *dp, uint64_t txg) | |
91 | { | |
92 | dmu_tx_t *tx = dmu_tx_create_dd(NULL); | |
93 | ||
94 | ASSERT3U(txg, <=, dp->dp_tx.tx_open_txg); | |
95 | tx->tx_pool = dp; | |
96 | tx->tx_txg = txg; | |
97 | tx->tx_anyobj = TRUE; | |
98 | ||
99 | return (tx); | |
100 | } | |
101 | ||
102 | int | |
103 | dmu_tx_is_syncing(dmu_tx_t *tx) | |
104 | { | |
105 | return (tx->tx_anyobj); | |
106 | } | |
107 | ||
108 | int | |
109 | dmu_tx_private_ok(dmu_tx_t *tx) | |
110 | { | |
111 | return (tx->tx_anyobj); | |
112 | } | |
113 | ||
114 | static dmu_tx_hold_t * | |
115 | dmu_tx_hold_object_impl(dmu_tx_t *tx, objset_t *os, uint64_t object, | |
116 | enum dmu_tx_hold_type type, uint64_t arg1, uint64_t arg2) | |
117 | { | |
118 | dmu_tx_hold_t *txh; | |
119 | dnode_t *dn = NULL; | |
120 | int err; | |
121 | ||
122 | if (object != DMU_NEW_OBJECT) { | |
428870ff | 123 | err = dnode_hold(os, object, tx, &dn); |
34dc7c2f BB |
124 | if (err) { |
125 | tx->tx_err = err; | |
126 | return (NULL); | |
127 | } | |
128 | ||
129 | if (err == 0 && tx->tx_txg != 0) { | |
130 | mutex_enter(&dn->dn_mtx); | |
131 | /* | |
132 | * dn->dn_assigned_txg == tx->tx_txg doesn't pose a | |
133 | * problem, but there's no way for it to happen (for | |
134 | * now, at least). | |
135 | */ | |
136 | ASSERT(dn->dn_assigned_txg == 0); | |
137 | dn->dn_assigned_txg = tx->tx_txg; | |
138 | (void) refcount_add(&dn->dn_tx_holds, tx); | |
139 | mutex_exit(&dn->dn_mtx); | |
140 | } | |
141 | } | |
142 | ||
b8d06fca | 143 | txh = kmem_zalloc(sizeof (dmu_tx_hold_t), KM_PUSHPAGE); |
34dc7c2f BB |
144 | txh->txh_tx = tx; |
145 | txh->txh_dnode = dn; | |
1c5de20a | 146 | #ifdef DEBUG_DMU_TX |
34dc7c2f BB |
147 | txh->txh_type = type; |
148 | txh->txh_arg1 = arg1; | |
149 | txh->txh_arg2 = arg2; | |
150 | #endif | |
151 | list_insert_tail(&tx->tx_holds, txh); | |
152 | ||
153 | return (txh); | |
154 | } | |
155 | ||
156 | void | |
157 | dmu_tx_add_new_object(dmu_tx_t *tx, objset_t *os, uint64_t object) | |
158 | { | |
159 | /* | |
160 | * If we're syncing, they can manipulate any object anyhow, and | |
161 | * the hold on the dnode_t can cause problems. | |
162 | */ | |
163 | if (!dmu_tx_is_syncing(tx)) { | |
164 | (void) dmu_tx_hold_object_impl(tx, os, | |
165 | object, THT_NEWOBJECT, 0, 0); | |
166 | } | |
167 | } | |
168 | ||
169 | static int | |
170 | dmu_tx_check_ioerr(zio_t *zio, dnode_t *dn, int level, uint64_t blkid) | |
171 | { | |
172 | int err; | |
173 | dmu_buf_impl_t *db; | |
174 | ||
175 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
176 | db = dbuf_hold_level(dn, level, blkid, FTAG); | |
177 | rw_exit(&dn->dn_struct_rwlock); | |
178 | if (db == NULL) | |
179 | return (EIO); | |
180 | err = dbuf_read(db, zio, DB_RF_CANFAIL | DB_RF_NOPREFETCH); | |
181 | dbuf_rele(db, FTAG); | |
182 | return (err); | |
183 | } | |
184 | ||
9babb374 | 185 | static void |
428870ff BB |
186 | dmu_tx_count_twig(dmu_tx_hold_t *txh, dnode_t *dn, dmu_buf_impl_t *db, |
187 | int level, uint64_t blkid, boolean_t freeable, uint64_t *history) | |
9babb374 | 188 | { |
428870ff BB |
189 | objset_t *os = dn->dn_objset; |
190 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
191 | int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
192 | dmu_buf_impl_t *parent = NULL; | |
193 | blkptr_t *bp = NULL; | |
194 | uint64_t space; | |
195 | ||
196 | if (level >= dn->dn_nlevels || history[level] == blkid) | |
9babb374 BB |
197 | return; |
198 | ||
428870ff | 199 | history[level] = blkid; |
9babb374 | 200 | |
428870ff BB |
201 | space = (level == 0) ? dn->dn_datablksz : (1ULL << dn->dn_indblkshift); |
202 | ||
203 | if (db == NULL || db == dn->dn_dbuf) { | |
204 | ASSERT(level != 0); | |
205 | db = NULL; | |
206 | } else { | |
572e2857 | 207 | ASSERT(DB_DNODE(db) == dn); |
428870ff BB |
208 | ASSERT(db->db_level == level); |
209 | ASSERT(db->db.db_size == space); | |
210 | ASSERT(db->db_blkid == blkid); | |
211 | bp = db->db_blkptr; | |
212 | parent = db->db_parent; | |
9babb374 BB |
213 | } |
214 | ||
428870ff BB |
215 | freeable = (bp && (freeable || |
216 | dsl_dataset_block_freeable(ds, bp, bp->blk_birth))); | |
9babb374 | 217 | |
428870ff BB |
218 | if (freeable) |
219 | txh->txh_space_tooverwrite += space; | |
220 | else | |
221 | txh->txh_space_towrite += space; | |
222 | if (bp) | |
223 | txh->txh_space_tounref += bp_get_dsize(os->os_spa, bp); | |
224 | ||
225 | dmu_tx_count_twig(txh, dn, parent, level + 1, | |
226 | blkid >> epbs, freeable, history); | |
9babb374 BB |
227 | } |
228 | ||
34dc7c2f BB |
229 | /* ARGSUSED */ |
230 | static void | |
231 | dmu_tx_count_write(dmu_tx_hold_t *txh, uint64_t off, uint64_t len) | |
232 | { | |
233 | dnode_t *dn = txh->txh_dnode; | |
234 | uint64_t start, end, i; | |
235 | int min_bs, max_bs, min_ibs, max_ibs, epbs, bits; | |
236 | int err = 0; | |
d6320ddb | 237 | int l; |
34dc7c2f BB |
238 | |
239 | if (len == 0) | |
240 | return; | |
241 | ||
242 | min_bs = SPA_MINBLOCKSHIFT; | |
243 | max_bs = SPA_MAXBLOCKSHIFT; | |
244 | min_ibs = DN_MIN_INDBLKSHIFT; | |
245 | max_ibs = DN_MAX_INDBLKSHIFT; | |
246 | ||
34dc7c2f | 247 | if (dn) { |
428870ff | 248 | uint64_t history[DN_MAX_LEVELS]; |
9babb374 BB |
249 | int nlvls = dn->dn_nlevels; |
250 | int delta; | |
251 | ||
252 | /* | |
253 | * For i/o error checking, read the first and last level-0 | |
254 | * blocks (if they are not aligned), and all the level-1 blocks. | |
255 | */ | |
34dc7c2f | 256 | if (dn->dn_maxblkid == 0) { |
9babb374 BB |
257 | delta = dn->dn_datablksz; |
258 | start = (off < dn->dn_datablksz) ? 0 : 1; | |
259 | end = (off+len <= dn->dn_datablksz) ? 0 : 1; | |
260 | if (start == 0 && (off > 0 || len < dn->dn_datablksz)) { | |
b128c09f BB |
261 | err = dmu_tx_check_ioerr(NULL, dn, 0, 0); |
262 | if (err) | |
263 | goto out; | |
9babb374 | 264 | delta -= off; |
b128c09f | 265 | } |
34dc7c2f BB |
266 | } else { |
267 | zio_t *zio = zio_root(dn->dn_objset->os_spa, | |
268 | NULL, NULL, ZIO_FLAG_CANFAIL); | |
269 | ||
270 | /* first level-0 block */ | |
271 | start = off >> dn->dn_datablkshift; | |
272 | if (P2PHASE(off, dn->dn_datablksz) || | |
273 | len < dn->dn_datablksz) { | |
274 | err = dmu_tx_check_ioerr(zio, dn, 0, start); | |
275 | if (err) | |
276 | goto out; | |
277 | } | |
278 | ||
279 | /* last level-0 block */ | |
280 | end = (off+len-1) >> dn->dn_datablkshift; | |
b128c09f | 281 | if (end != start && end <= dn->dn_maxblkid && |
34dc7c2f BB |
282 | P2PHASE(off+len, dn->dn_datablksz)) { |
283 | err = dmu_tx_check_ioerr(zio, dn, 0, end); | |
284 | if (err) | |
285 | goto out; | |
286 | } | |
287 | ||
288 | /* level-1 blocks */ | |
9babb374 BB |
289 | if (nlvls > 1) { |
290 | int shft = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
291 | for (i = (start>>shft)+1; i < end>>shft; i++) { | |
34dc7c2f BB |
292 | err = dmu_tx_check_ioerr(zio, dn, 1, i); |
293 | if (err) | |
294 | goto out; | |
295 | } | |
296 | } | |
297 | ||
298 | err = zio_wait(zio); | |
299 | if (err) | |
300 | goto out; | |
9babb374 | 301 | delta = P2NPHASE(off, dn->dn_datablksz); |
34dc7c2f | 302 | } |
34dc7c2f | 303 | |
295304be | 304 | min_ibs = max_ibs = dn->dn_indblkshift; |
9babb374 BB |
305 | if (dn->dn_maxblkid > 0) { |
306 | /* | |
307 | * The blocksize can't change, | |
308 | * so we can make a more precise estimate. | |
309 | */ | |
310 | ASSERT(dn->dn_datablkshift != 0); | |
34dc7c2f | 311 | min_bs = max_bs = dn->dn_datablkshift; |
9babb374 BB |
312 | } |
313 | ||
314 | /* | |
315 | * If this write is not off the end of the file | |
316 | * we need to account for overwrites/unref. | |
317 | */ | |
428870ff | 318 | if (start <= dn->dn_maxblkid) { |
d6320ddb | 319 | for (l = 0; l < DN_MAX_LEVELS; l++) |
428870ff BB |
320 | history[l] = -1ULL; |
321 | } | |
9babb374 | 322 | while (start <= dn->dn_maxblkid) { |
9babb374 BB |
323 | dmu_buf_impl_t *db; |
324 | ||
325 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
428870ff | 326 | err = dbuf_hold_impl(dn, 0, start, FALSE, FTAG, &db); |
9babb374 | 327 | rw_exit(&dn->dn_struct_rwlock); |
428870ff BB |
328 | |
329 | if (err) { | |
330 | txh->txh_tx->tx_err = err; | |
331 | return; | |
9babb374 | 332 | } |
428870ff BB |
333 | |
334 | dmu_tx_count_twig(txh, dn, db, 0, start, B_FALSE, | |
335 | history); | |
9babb374 BB |
336 | dbuf_rele(db, FTAG); |
337 | if (++start > end) { | |
338 | /* | |
339 | * Account for new indirects appearing | |
340 | * before this IO gets assigned into a txg. | |
341 | */ | |
342 | bits = 64 - min_bs; | |
343 | epbs = min_ibs - SPA_BLKPTRSHIFT; | |
344 | for (bits -= epbs * (nlvls - 1); | |
345 | bits >= 0; bits -= epbs) | |
346 | txh->txh_fudge += 1ULL << max_ibs; | |
347 | goto out; | |
348 | } | |
349 | off += delta; | |
350 | if (len >= delta) | |
351 | len -= delta; | |
352 | delta = dn->dn_datablksz; | |
353 | } | |
34dc7c2f BB |
354 | } |
355 | ||
356 | /* | |
357 | * 'end' is the last thing we will access, not one past. | |
358 | * This way we won't overflow when accessing the last byte. | |
359 | */ | |
360 | start = P2ALIGN(off, 1ULL << max_bs); | |
361 | end = P2ROUNDUP(off + len, 1ULL << max_bs) - 1; | |
362 | txh->txh_space_towrite += end - start + 1; | |
363 | ||
364 | start >>= min_bs; | |
365 | end >>= min_bs; | |
366 | ||
367 | epbs = min_ibs - SPA_BLKPTRSHIFT; | |
368 | ||
369 | /* | |
370 | * The object contains at most 2^(64 - min_bs) blocks, | |
371 | * and each indirect level maps 2^epbs. | |
372 | */ | |
373 | for (bits = 64 - min_bs; bits >= 0; bits -= epbs) { | |
374 | start >>= epbs; | |
375 | end >>= epbs; | |
9babb374 | 376 | ASSERT3U(end, >=, start); |
34dc7c2f | 377 | txh->txh_space_towrite += (end - start + 1) << max_ibs; |
9babb374 BB |
378 | if (start != 0) { |
379 | /* | |
380 | * We also need a new blkid=0 indirect block | |
381 | * to reference any existing file data. | |
382 | */ | |
383 | txh->txh_space_towrite += 1ULL << max_ibs; | |
384 | } | |
34dc7c2f BB |
385 | } |
386 | ||
34dc7c2f | 387 | out: |
9babb374 BB |
388 | if (txh->txh_space_towrite + txh->txh_space_tooverwrite > |
389 | 2 * DMU_MAX_ACCESS) | |
390 | err = EFBIG; | |
391 | ||
34dc7c2f BB |
392 | if (err) |
393 | txh->txh_tx->tx_err = err; | |
394 | } | |
395 | ||
396 | static void | |
397 | dmu_tx_count_dnode(dmu_tx_hold_t *txh) | |
398 | { | |
399 | dnode_t *dn = txh->txh_dnode; | |
572e2857 | 400 | dnode_t *mdn = DMU_META_DNODE(txh->txh_tx->tx_objset); |
34dc7c2f BB |
401 | uint64_t space = mdn->dn_datablksz + |
402 | ((mdn->dn_nlevels-1) << mdn->dn_indblkshift); | |
403 | ||
404 | if (dn && dn->dn_dbuf->db_blkptr && | |
405 | dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset, | |
428870ff | 406 | dn->dn_dbuf->db_blkptr, dn->dn_dbuf->db_blkptr->blk_birth)) { |
34dc7c2f | 407 | txh->txh_space_tooverwrite += space; |
9babb374 | 408 | txh->txh_space_tounref += space; |
34dc7c2f BB |
409 | } else { |
410 | txh->txh_space_towrite += space; | |
411 | if (dn && dn->dn_dbuf->db_blkptr) | |
412 | txh->txh_space_tounref += space; | |
413 | } | |
414 | } | |
415 | ||
416 | void | |
417 | dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len) | |
418 | { | |
419 | dmu_tx_hold_t *txh; | |
420 | ||
421 | ASSERT(tx->tx_txg == 0); | |
422 | ASSERT(len < DMU_MAX_ACCESS); | |
423 | ASSERT(len == 0 || UINT64_MAX - off >= len - 1); | |
424 | ||
425 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, | |
426 | object, THT_WRITE, off, len); | |
427 | if (txh == NULL) | |
428 | return; | |
429 | ||
430 | dmu_tx_count_write(txh, off, len); | |
431 | dmu_tx_count_dnode(txh); | |
432 | } | |
433 | ||
434 | static void | |
435 | dmu_tx_count_free(dmu_tx_hold_t *txh, uint64_t off, uint64_t len) | |
436 | { | |
b128c09f BB |
437 | uint64_t blkid, nblks, lastblk; |
438 | uint64_t space = 0, unref = 0, skipped = 0; | |
34dc7c2f BB |
439 | dnode_t *dn = txh->txh_dnode; |
440 | dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; | |
441 | spa_t *spa = txh->txh_tx->tx_pool->dp_spa; | |
b128c09f | 442 | int epbs; |
ff80d9b1 | 443 | uint64_t l0span = 0, nl1blks = 0; |
34dc7c2f | 444 | |
b128c09f | 445 | if (dn->dn_nlevels == 0) |
34dc7c2f BB |
446 | return; |
447 | ||
448 | /* | |
b128c09f | 449 | * The struct_rwlock protects us against dn_nlevels |
34dc7c2f BB |
450 | * changing, in case (against all odds) we manage to dirty & |
451 | * sync out the changes after we check for being dirty. | |
428870ff | 452 | * Also, dbuf_hold_impl() wants us to have the struct_rwlock. |
34dc7c2f BB |
453 | */ |
454 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
b128c09f BB |
455 | epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; |
456 | if (dn->dn_maxblkid == 0) { | |
34dc7c2f BB |
457 | if (off == 0 && len >= dn->dn_datablksz) { |
458 | blkid = 0; | |
459 | nblks = 1; | |
460 | } else { | |
461 | rw_exit(&dn->dn_struct_rwlock); | |
462 | return; | |
463 | } | |
464 | } else { | |
465 | blkid = off >> dn->dn_datablkshift; | |
b128c09f | 466 | nblks = (len + dn->dn_datablksz - 1) >> dn->dn_datablkshift; |
34dc7c2f | 467 | |
b128c09f | 468 | if (blkid >= dn->dn_maxblkid) { |
34dc7c2f BB |
469 | rw_exit(&dn->dn_struct_rwlock); |
470 | return; | |
471 | } | |
b128c09f BB |
472 | if (blkid + nblks > dn->dn_maxblkid) |
473 | nblks = dn->dn_maxblkid - blkid; | |
34dc7c2f | 474 | |
34dc7c2f | 475 | } |
ff80d9b1 | 476 | l0span = nblks; /* save for later use to calc level > 1 overhead */ |
b128c09f | 477 | if (dn->dn_nlevels == 1) { |
34dc7c2f BB |
478 | int i; |
479 | for (i = 0; i < nblks; i++) { | |
480 | blkptr_t *bp = dn->dn_phys->dn_blkptr; | |
b128c09f | 481 | ASSERT3U(blkid + i, <, dn->dn_nblkptr); |
34dc7c2f | 482 | bp += blkid + i; |
428870ff | 483 | if (dsl_dataset_block_freeable(ds, bp, bp->blk_birth)) { |
34dc7c2f | 484 | dprintf_bp(bp, "can free old%s", ""); |
428870ff | 485 | space += bp_get_dsize(spa, bp); |
34dc7c2f BB |
486 | } |
487 | unref += BP_GET_ASIZE(bp); | |
488 | } | |
ff80d9b1 | 489 | nl1blks = 1; |
34dc7c2f BB |
490 | nblks = 0; |
491 | } | |
492 | ||
b128c09f | 493 | lastblk = blkid + nblks - 1; |
34dc7c2f BB |
494 | while (nblks) { |
495 | dmu_buf_impl_t *dbuf; | |
b128c09f BB |
496 | uint64_t ibyte, new_blkid; |
497 | int epb = 1 << epbs; | |
498 | int err, i, blkoff, tochk; | |
499 | blkptr_t *bp; | |
500 | ||
501 | ibyte = blkid << dn->dn_datablkshift; | |
502 | err = dnode_next_offset(dn, | |
503 | DNODE_FIND_HAVELOCK, &ibyte, 2, 1, 0); | |
504 | new_blkid = ibyte >> dn->dn_datablkshift; | |
505 | if (err == ESRCH) { | |
506 | skipped += (lastblk >> epbs) - (blkid >> epbs) + 1; | |
507 | break; | |
508 | } | |
509 | if (err) { | |
510 | txh->txh_tx->tx_err = err; | |
511 | break; | |
512 | } | |
513 | if (new_blkid > lastblk) { | |
514 | skipped += (lastblk >> epbs) - (blkid >> epbs) + 1; | |
515 | break; | |
516 | } | |
34dc7c2f | 517 | |
b128c09f BB |
518 | if (new_blkid > blkid) { |
519 | ASSERT((new_blkid >> epbs) > (blkid >> epbs)); | |
520 | skipped += (new_blkid >> epbs) - (blkid >> epbs) - 1; | |
521 | nblks -= new_blkid - blkid; | |
522 | blkid = new_blkid; | |
523 | } | |
524 | blkoff = P2PHASE(blkid, epb); | |
525 | tochk = MIN(epb - blkoff, nblks); | |
34dc7c2f | 526 | |
428870ff BB |
527 | err = dbuf_hold_impl(dn, 1, blkid >> epbs, FALSE, FTAG, &dbuf); |
528 | if (err) { | |
529 | txh->txh_tx->tx_err = err; | |
b128c09f | 530 | break; |
34dc7c2f | 531 | } |
428870ff BB |
532 | |
533 | txh->txh_memory_tohold += dbuf->db.db_size; | |
534 | ||
535 | /* | |
536 | * We don't check memory_tohold against DMU_MAX_ACCESS because | |
537 | * memory_tohold is an over-estimation (especially the >L1 | |
538 | * indirect blocks), so it could fail. Callers should have | |
539 | * already verified that they will not be holding too much | |
540 | * memory. | |
541 | */ | |
542 | ||
b128c09f BB |
543 | err = dbuf_read(dbuf, NULL, DB_RF_HAVESTRUCT | DB_RF_CANFAIL); |
544 | if (err != 0) { | |
34dc7c2f | 545 | txh->txh_tx->tx_err = err; |
b128c09f | 546 | dbuf_rele(dbuf, FTAG); |
34dc7c2f BB |
547 | break; |
548 | } | |
549 | ||
b128c09f BB |
550 | bp = dbuf->db.db_data; |
551 | bp += blkoff; | |
552 | ||
553 | for (i = 0; i < tochk; i++) { | |
428870ff BB |
554 | if (dsl_dataset_block_freeable(ds, &bp[i], |
555 | bp[i].blk_birth)) { | |
b128c09f | 556 | dprintf_bp(&bp[i], "can free old%s", ""); |
428870ff | 557 | space += bp_get_dsize(spa, &bp[i]); |
b128c09f BB |
558 | } |
559 | unref += BP_GET_ASIZE(bp); | |
560 | } | |
561 | dbuf_rele(dbuf, FTAG); | |
562 | ||
ff80d9b1 | 563 | ++nl1blks; |
34dc7c2f BB |
564 | blkid += tochk; |
565 | nblks -= tochk; | |
566 | } | |
567 | rw_exit(&dn->dn_struct_rwlock); | |
568 | ||
ff80d9b1 AJ |
569 | /* |
570 | * Add in memory requirements of higher-level indirects. | |
571 | * This assumes a worst-possible scenario for dn_nlevels and a | |
572 | * worst-possible distribution of l1-blocks over the region to free. | |
573 | */ | |
574 | { | |
575 | uint64_t blkcnt = 1 + ((l0span >> epbs) >> epbs); | |
576 | int level = 2; | |
577 | /* | |
578 | * Here we don't use DN_MAX_LEVEL, but calculate it with the | |
579 | * given datablkshift and indblkshift. This makes the | |
580 | * difference between 19 and 8 on large files. | |
581 | */ | |
582 | int maxlevel = 2 + (DN_MAX_OFFSET_SHIFT - dn->dn_datablkshift) / | |
583 | (dn->dn_indblkshift - SPA_BLKPTRSHIFT); | |
584 | ||
585 | while (level++ < maxlevel) { | |
b077fd4c | 586 | txh->txh_memory_tohold += MAX(MIN(blkcnt, nl1blks), 1) |
ff80d9b1 AJ |
587 | << dn->dn_indblkshift; |
588 | blkcnt = 1 + (blkcnt >> epbs); | |
589 | } | |
590 | } | |
591 | ||
b128c09f BB |
592 | /* account for new level 1 indirect blocks that might show up */ |
593 | if (skipped > 0) { | |
594 | txh->txh_fudge += skipped << dn->dn_indblkshift; | |
595 | skipped = MIN(skipped, DMU_MAX_DELETEBLKCNT >> epbs); | |
596 | txh->txh_memory_tohold += skipped << dn->dn_indblkshift; | |
597 | } | |
34dc7c2f BB |
598 | txh->txh_space_tofree += space; |
599 | txh->txh_space_tounref += unref; | |
600 | } | |
601 | ||
602 | void | |
603 | dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, uint64_t len) | |
604 | { | |
605 | dmu_tx_hold_t *txh; | |
606 | dnode_t *dn; | |
607 | uint64_t start, end, i; | |
608 | int err, shift; | |
609 | zio_t *zio; | |
610 | ||
611 | ASSERT(tx->tx_txg == 0); | |
612 | ||
613 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, | |
614 | object, THT_FREE, off, len); | |
615 | if (txh == NULL) | |
616 | return; | |
617 | dn = txh->txh_dnode; | |
618 | ||
619 | /* first block */ | |
620 | if (off != 0) | |
621 | dmu_tx_count_write(txh, off, 1); | |
622 | /* last block */ | |
623 | if (len != DMU_OBJECT_END) | |
624 | dmu_tx_count_write(txh, off+len, 1); | |
625 | ||
428870ff BB |
626 | dmu_tx_count_dnode(txh); |
627 | ||
34dc7c2f BB |
628 | if (off >= (dn->dn_maxblkid+1) * dn->dn_datablksz) |
629 | return; | |
630 | if (len == DMU_OBJECT_END) | |
631 | len = (dn->dn_maxblkid+1) * dn->dn_datablksz - off; | |
632 | ||
633 | /* | |
634 | * For i/o error checking, read the first and last level-0 | |
635 | * blocks, and all the level-1 blocks. The above count_write's | |
b128c09f | 636 | * have already taken care of the level-0 blocks. |
34dc7c2f BB |
637 | */ |
638 | if (dn->dn_nlevels > 1) { | |
639 | shift = dn->dn_datablkshift + dn->dn_indblkshift - | |
640 | SPA_BLKPTRSHIFT; | |
641 | start = off >> shift; | |
642 | end = dn->dn_datablkshift ? ((off+len) >> shift) : 0; | |
643 | ||
644 | zio = zio_root(tx->tx_pool->dp_spa, | |
645 | NULL, NULL, ZIO_FLAG_CANFAIL); | |
646 | for (i = start; i <= end; i++) { | |
647 | uint64_t ibyte = i << shift; | |
b128c09f | 648 | err = dnode_next_offset(dn, 0, &ibyte, 2, 1, 0); |
34dc7c2f BB |
649 | i = ibyte >> shift; |
650 | if (err == ESRCH) | |
651 | break; | |
652 | if (err) { | |
653 | tx->tx_err = err; | |
654 | return; | |
655 | } | |
656 | ||
657 | err = dmu_tx_check_ioerr(zio, dn, 1, i); | |
658 | if (err) { | |
659 | tx->tx_err = err; | |
660 | return; | |
661 | } | |
662 | } | |
663 | err = zio_wait(zio); | |
664 | if (err) { | |
665 | tx->tx_err = err; | |
666 | return; | |
667 | } | |
668 | } | |
669 | ||
34dc7c2f BB |
670 | dmu_tx_count_free(txh, off, len); |
671 | } | |
672 | ||
673 | void | |
9babb374 | 674 | dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name) |
34dc7c2f BB |
675 | { |
676 | dmu_tx_hold_t *txh; | |
677 | dnode_t *dn; | |
678 | uint64_t nblocks; | |
679 | int epbs, err; | |
680 | ||
681 | ASSERT(tx->tx_txg == 0); | |
682 | ||
683 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, | |
684 | object, THT_ZAP, add, (uintptr_t)name); | |
685 | if (txh == NULL) | |
686 | return; | |
687 | dn = txh->txh_dnode; | |
688 | ||
689 | dmu_tx_count_dnode(txh); | |
690 | ||
691 | if (dn == NULL) { | |
692 | /* | |
693 | * We will be able to fit a new object's entries into one leaf | |
694 | * block. So there will be at most 2 blocks total, | |
695 | * including the header block. | |
696 | */ | |
697 | dmu_tx_count_write(txh, 0, 2 << fzap_default_block_shift); | |
698 | return; | |
699 | } | |
700 | ||
9ae529ec | 701 | ASSERT3U(DMU_OT_BYTESWAP(dn->dn_type), ==, DMU_BSWAP_ZAP); |
34dc7c2f BB |
702 | |
703 | if (dn->dn_maxblkid == 0 && !add) { | |
22cd4a46 AL |
704 | blkptr_t *bp; |
705 | ||
34dc7c2f BB |
706 | /* |
707 | * If there is only one block (i.e. this is a micro-zap) | |
708 | * and we are not adding anything, the accounting is simple. | |
709 | */ | |
710 | err = dmu_tx_check_ioerr(NULL, dn, 0, 0); | |
711 | if (err) { | |
712 | tx->tx_err = err; | |
713 | return; | |
714 | } | |
715 | ||
716 | /* | |
717 | * Use max block size here, since we don't know how much | |
718 | * the size will change between now and the dbuf dirty call. | |
719 | */ | |
22cd4a46 | 720 | bp = &dn->dn_phys->dn_blkptr[0]; |
34dc7c2f | 721 | if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset, |
22cd4a46 | 722 | bp, bp->blk_birth)) |
34dc7c2f | 723 | txh->txh_space_tooverwrite += SPA_MAXBLOCKSIZE; |
22cd4a46 | 724 | else |
34dc7c2f | 725 | txh->txh_space_towrite += SPA_MAXBLOCKSIZE; |
22cd4a46 | 726 | if (!BP_IS_HOLE(bp)) |
9babb374 | 727 | txh->txh_space_tounref += SPA_MAXBLOCKSIZE; |
34dc7c2f BB |
728 | return; |
729 | } | |
730 | ||
731 | if (dn->dn_maxblkid > 0 && name) { | |
732 | /* | |
733 | * access the name in this fat-zap so that we'll check | |
734 | * for i/o errors to the leaf blocks, etc. | |
735 | */ | |
428870ff | 736 | err = zap_lookup(dn->dn_objset, dn->dn_object, name, |
34dc7c2f BB |
737 | 8, 0, NULL); |
738 | if (err == EIO) { | |
739 | tx->tx_err = err; | |
740 | return; | |
741 | } | |
742 | } | |
743 | ||
428870ff | 744 | err = zap_count_write(dn->dn_objset, dn->dn_object, name, add, |
45d1cae3 | 745 | &txh->txh_space_towrite, &txh->txh_space_tooverwrite); |
34dc7c2f BB |
746 | |
747 | /* | |
748 | * If the modified blocks are scattered to the four winds, | |
749 | * we'll have to modify an indirect twig for each. | |
750 | */ | |
751 | epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
752 | for (nblocks = dn->dn_maxblkid >> epbs; nblocks != 0; nblocks >>= epbs) | |
9babb374 BB |
753 | if (dn->dn_objset->os_dsl_dataset->ds_phys->ds_prev_snap_obj) |
754 | txh->txh_space_towrite += 3 << dn->dn_indblkshift; | |
755 | else | |
756 | txh->txh_space_tooverwrite += 3 << dn->dn_indblkshift; | |
34dc7c2f BB |
757 | } |
758 | ||
759 | void | |
760 | dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object) | |
761 | { | |
762 | dmu_tx_hold_t *txh; | |
763 | ||
764 | ASSERT(tx->tx_txg == 0); | |
765 | ||
766 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, | |
767 | object, THT_BONUS, 0, 0); | |
768 | if (txh) | |
769 | dmu_tx_count_dnode(txh); | |
770 | } | |
771 | ||
772 | void | |
773 | dmu_tx_hold_space(dmu_tx_t *tx, uint64_t space) | |
774 | { | |
775 | dmu_tx_hold_t *txh; | |
776 | ASSERT(tx->tx_txg == 0); | |
777 | ||
778 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, | |
779 | DMU_NEW_OBJECT, THT_SPACE, space, 0); | |
780 | ||
781 | txh->txh_space_towrite += space; | |
782 | } | |
783 | ||
784 | int | |
785 | dmu_tx_holds(dmu_tx_t *tx, uint64_t object) | |
786 | { | |
787 | dmu_tx_hold_t *txh; | |
788 | int holds = 0; | |
789 | ||
790 | /* | |
791 | * By asserting that the tx is assigned, we're counting the | |
792 | * number of dn_tx_holds, which is the same as the number of | |
793 | * dn_holds. Otherwise, we'd be counting dn_holds, but | |
794 | * dn_tx_holds could be 0. | |
795 | */ | |
796 | ASSERT(tx->tx_txg != 0); | |
797 | ||
798 | /* if (tx->tx_anyobj == TRUE) */ | |
799 | /* return (0); */ | |
800 | ||
801 | for (txh = list_head(&tx->tx_holds); txh; | |
802 | txh = list_next(&tx->tx_holds, txh)) { | |
803 | if (txh->txh_dnode && txh->txh_dnode->dn_object == object) | |
804 | holds++; | |
805 | } | |
806 | ||
807 | return (holds); | |
808 | } | |
809 | ||
1c5de20a | 810 | #ifdef DEBUG_DMU_TX |
34dc7c2f BB |
811 | void |
812 | dmu_tx_dirty_buf(dmu_tx_t *tx, dmu_buf_impl_t *db) | |
813 | { | |
814 | dmu_tx_hold_t *txh; | |
815 | int match_object = FALSE, match_offset = FALSE; | |
572e2857 | 816 | dnode_t *dn; |
34dc7c2f | 817 | |
572e2857 BB |
818 | DB_DNODE_ENTER(db); |
819 | dn = DB_DNODE(db); | |
99ea23c5 | 820 | ASSERT(dn != NULL); |
34dc7c2f | 821 | ASSERT(tx->tx_txg != 0); |
428870ff | 822 | ASSERT(tx->tx_objset == NULL || dn->dn_objset == tx->tx_objset); |
34dc7c2f BB |
823 | ASSERT3U(dn->dn_object, ==, db->db.db_object); |
824 | ||
572e2857 BB |
825 | if (tx->tx_anyobj) { |
826 | DB_DNODE_EXIT(db); | |
34dc7c2f | 827 | return; |
572e2857 | 828 | } |
34dc7c2f BB |
829 | |
830 | /* XXX No checking on the meta dnode for now */ | |
572e2857 BB |
831 | if (db->db.db_object == DMU_META_DNODE_OBJECT) { |
832 | DB_DNODE_EXIT(db); | |
34dc7c2f | 833 | return; |
572e2857 | 834 | } |
34dc7c2f BB |
835 | |
836 | for (txh = list_head(&tx->tx_holds); txh; | |
837 | txh = list_next(&tx->tx_holds, txh)) { | |
99ea23c5 | 838 | ASSERT3U(dn->dn_assigned_txg, ==, tx->tx_txg); |
34dc7c2f BB |
839 | if (txh->txh_dnode == dn && txh->txh_type != THT_NEWOBJECT) |
840 | match_object = TRUE; | |
841 | if (txh->txh_dnode == NULL || txh->txh_dnode == dn) { | |
842 | int datablkshift = dn->dn_datablkshift ? | |
843 | dn->dn_datablkshift : SPA_MAXBLOCKSHIFT; | |
844 | int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
845 | int shift = datablkshift + epbs * db->db_level; | |
846 | uint64_t beginblk = shift >= 64 ? 0 : | |
847 | (txh->txh_arg1 >> shift); | |
848 | uint64_t endblk = shift >= 64 ? 0 : | |
849 | ((txh->txh_arg1 + txh->txh_arg2 - 1) >> shift); | |
850 | uint64_t blkid = db->db_blkid; | |
851 | ||
852 | /* XXX txh_arg2 better not be zero... */ | |
853 | ||
854 | dprintf("found txh type %x beginblk=%llx endblk=%llx\n", | |
855 | txh->txh_type, beginblk, endblk); | |
856 | ||
857 | switch (txh->txh_type) { | |
858 | case THT_WRITE: | |
859 | if (blkid >= beginblk && blkid <= endblk) | |
860 | match_offset = TRUE; | |
861 | /* | |
862 | * We will let this hold work for the bonus | |
428870ff BB |
863 | * or spill buffer so that we don't need to |
864 | * hold it when creating a new object. | |
34dc7c2f | 865 | */ |
428870ff BB |
866 | if (blkid == DMU_BONUS_BLKID || |
867 | blkid == DMU_SPILL_BLKID) | |
34dc7c2f BB |
868 | match_offset = TRUE; |
869 | /* | |
870 | * They might have to increase nlevels, | |
871 | * thus dirtying the new TLIBs. Or the | |
872 | * might have to change the block size, | |
873 | * thus dirying the new lvl=0 blk=0. | |
874 | */ | |
875 | if (blkid == 0) | |
876 | match_offset = TRUE; | |
877 | break; | |
878 | case THT_FREE: | |
b128c09f BB |
879 | /* |
880 | * We will dirty all the level 1 blocks in | |
881 | * the free range and perhaps the first and | |
882 | * last level 0 block. | |
883 | */ | |
884 | if (blkid >= beginblk && (blkid <= endblk || | |
885 | txh->txh_arg2 == DMU_OBJECT_END)) | |
34dc7c2f BB |
886 | match_offset = TRUE; |
887 | break; | |
428870ff BB |
888 | case THT_SPILL: |
889 | if (blkid == DMU_SPILL_BLKID) | |
890 | match_offset = TRUE; | |
891 | break; | |
34dc7c2f | 892 | case THT_BONUS: |
428870ff | 893 | if (blkid == DMU_BONUS_BLKID) |
34dc7c2f BB |
894 | match_offset = TRUE; |
895 | break; | |
896 | case THT_ZAP: | |
897 | match_offset = TRUE; | |
898 | break; | |
899 | case THT_NEWOBJECT: | |
900 | match_object = TRUE; | |
901 | break; | |
902 | default: | |
903 | ASSERT(!"bad txh_type"); | |
904 | } | |
905 | } | |
572e2857 BB |
906 | if (match_object && match_offset) { |
907 | DB_DNODE_EXIT(db); | |
34dc7c2f | 908 | return; |
572e2857 | 909 | } |
34dc7c2f | 910 | } |
572e2857 | 911 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
912 | panic("dirtying dbuf obj=%llx lvl=%u blkid=%llx but not tx_held\n", |
913 | (u_longlong_t)db->db.db_object, db->db_level, | |
914 | (u_longlong_t)db->db_blkid); | |
915 | } | |
916 | #endif | |
917 | ||
918 | static int | |
919 | dmu_tx_try_assign(dmu_tx_t *tx, uint64_t txg_how) | |
920 | { | |
921 | dmu_tx_hold_t *txh; | |
922 | spa_t *spa = tx->tx_pool->dp_spa; | |
b128c09f BB |
923 | uint64_t memory, asize, fsize, usize; |
924 | uint64_t towrite, tofree, tooverwrite, tounref, tohold, fudge; | |
34dc7c2f | 925 | |
c99c9001 | 926 | ASSERT0(tx->tx_txg); |
34dc7c2f | 927 | |
570827e1 BB |
928 | if (tx->tx_err) { |
929 | DMU_TX_STAT_BUMP(dmu_tx_error); | |
34dc7c2f | 930 | return (tx->tx_err); |
570827e1 | 931 | } |
34dc7c2f | 932 | |
b128c09f | 933 | if (spa_suspended(spa)) { |
570827e1 BB |
934 | DMU_TX_STAT_BUMP(dmu_tx_suspended); |
935 | ||
34dc7c2f BB |
936 | /* |
937 | * If the user has indicated a blocking failure mode | |
938 | * then return ERESTART which will block in dmu_tx_wait(). | |
939 | * Otherwise, return EIO so that an error can get | |
940 | * propagated back to the VOP calls. | |
941 | * | |
942 | * Note that we always honor the txg_how flag regardless | |
943 | * of the failuremode setting. | |
944 | */ | |
945 | if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE && | |
946 | txg_how != TXG_WAIT) | |
947 | return (EIO); | |
948 | ||
949 | return (ERESTART); | |
950 | } | |
951 | ||
952 | tx->tx_txg = txg_hold_open(tx->tx_pool, &tx->tx_txgh); | |
953 | tx->tx_needassign_txh = NULL; | |
954 | ||
955 | /* | |
956 | * NB: No error returns are allowed after txg_hold_open, but | |
957 | * before processing the dnode holds, due to the | |
958 | * dmu_tx_unassign() logic. | |
959 | */ | |
960 | ||
b128c09f | 961 | towrite = tofree = tooverwrite = tounref = tohold = fudge = 0; |
34dc7c2f BB |
962 | for (txh = list_head(&tx->tx_holds); txh; |
963 | txh = list_next(&tx->tx_holds, txh)) { | |
964 | dnode_t *dn = txh->txh_dnode; | |
965 | if (dn != NULL) { | |
966 | mutex_enter(&dn->dn_mtx); | |
967 | if (dn->dn_assigned_txg == tx->tx_txg - 1) { | |
968 | mutex_exit(&dn->dn_mtx); | |
969 | tx->tx_needassign_txh = txh; | |
570827e1 | 970 | DMU_TX_STAT_BUMP(dmu_tx_group); |
34dc7c2f BB |
971 | return (ERESTART); |
972 | } | |
973 | if (dn->dn_assigned_txg == 0) | |
974 | dn->dn_assigned_txg = tx->tx_txg; | |
975 | ASSERT3U(dn->dn_assigned_txg, ==, tx->tx_txg); | |
976 | (void) refcount_add(&dn->dn_tx_holds, tx); | |
977 | mutex_exit(&dn->dn_mtx); | |
978 | } | |
979 | towrite += txh->txh_space_towrite; | |
980 | tofree += txh->txh_space_tofree; | |
981 | tooverwrite += txh->txh_space_tooverwrite; | |
982 | tounref += txh->txh_space_tounref; | |
b128c09f BB |
983 | tohold += txh->txh_memory_tohold; |
984 | fudge += txh->txh_fudge; | |
34dc7c2f BB |
985 | } |
986 | ||
987 | /* | |
988 | * NB: This check must be after we've held the dnodes, so that | |
989 | * the dmu_tx_unassign() logic will work properly | |
990 | */ | |
570827e1 BB |
991 | if (txg_how >= TXG_INITIAL && txg_how != tx->tx_txg) { |
992 | DMU_TX_STAT_BUMP(dmu_tx_how); | |
34dc7c2f | 993 | return (ERESTART); |
570827e1 | 994 | } |
34dc7c2f BB |
995 | |
996 | /* | |
997 | * If a snapshot has been taken since we made our estimates, | |
998 | * assume that we won't be able to free or overwrite anything. | |
999 | */ | |
1000 | if (tx->tx_objset && | |
428870ff | 1001 | dsl_dataset_prev_snap_txg(tx->tx_objset->os_dsl_dataset) > |
34dc7c2f BB |
1002 | tx->tx_lastsnap_txg) { |
1003 | towrite += tooverwrite; | |
1004 | tooverwrite = tofree = 0; | |
1005 | } | |
1006 | ||
b128c09f BB |
1007 | /* needed allocation: worst-case estimate of write space */ |
1008 | asize = spa_get_asize(tx->tx_pool->dp_spa, towrite + tooverwrite); | |
1009 | /* freed space estimate: worst-case overwrite + free estimate */ | |
34dc7c2f | 1010 | fsize = spa_get_asize(tx->tx_pool->dp_spa, tooverwrite) + tofree; |
b128c09f | 1011 | /* convert unrefd space to worst-case estimate */ |
34dc7c2f | 1012 | usize = spa_get_asize(tx->tx_pool->dp_spa, tounref); |
b128c09f BB |
1013 | /* calculate memory footprint estimate */ |
1014 | memory = towrite + tooverwrite + tohold; | |
34dc7c2f | 1015 | |
1c5de20a | 1016 | #ifdef DEBUG_DMU_TX |
b128c09f BB |
1017 | /* |
1018 | * Add in 'tohold' to account for our dirty holds on this memory | |
1019 | * XXX - the "fudge" factor is to account for skipped blocks that | |
1020 | * we missed because dnode_next_offset() misses in-core-only blocks. | |
1021 | */ | |
1022 | tx->tx_space_towrite = asize + | |
1023 | spa_get_asize(tx->tx_pool->dp_spa, tohold + fudge); | |
34dc7c2f BB |
1024 | tx->tx_space_tofree = tofree; |
1025 | tx->tx_space_tooverwrite = tooverwrite; | |
1026 | tx->tx_space_tounref = tounref; | |
1027 | #endif | |
1028 | ||
1029 | if (tx->tx_dir && asize != 0) { | |
b128c09f BB |
1030 | int err = dsl_dir_tempreserve_space(tx->tx_dir, memory, |
1031 | asize, fsize, usize, &tx->tx_tempreserve_cookie, tx); | |
34dc7c2f BB |
1032 | if (err) |
1033 | return (err); | |
1034 | } | |
1035 | ||
570827e1 BB |
1036 | DMU_TX_STAT_BUMP(dmu_tx_assigned); |
1037 | ||
34dc7c2f BB |
1038 | return (0); |
1039 | } | |
1040 | ||
1041 | static void | |
1042 | dmu_tx_unassign(dmu_tx_t *tx) | |
1043 | { | |
1044 | dmu_tx_hold_t *txh; | |
1045 | ||
1046 | if (tx->tx_txg == 0) | |
1047 | return; | |
1048 | ||
1049 | txg_rele_to_quiesce(&tx->tx_txgh); | |
1050 | ||
1051 | for (txh = list_head(&tx->tx_holds); txh != tx->tx_needassign_txh; | |
1052 | txh = list_next(&tx->tx_holds, txh)) { | |
1053 | dnode_t *dn = txh->txh_dnode; | |
1054 | ||
1055 | if (dn == NULL) | |
1056 | continue; | |
1057 | mutex_enter(&dn->dn_mtx); | |
1058 | ASSERT3U(dn->dn_assigned_txg, ==, tx->tx_txg); | |
1059 | ||
1060 | if (refcount_remove(&dn->dn_tx_holds, tx) == 0) { | |
1061 | dn->dn_assigned_txg = 0; | |
1062 | cv_broadcast(&dn->dn_notxholds); | |
1063 | } | |
1064 | mutex_exit(&dn->dn_mtx); | |
1065 | } | |
1066 | ||
1067 | txg_rele_to_sync(&tx->tx_txgh); | |
1068 | ||
1069 | tx->tx_lasttried_txg = tx->tx_txg; | |
1070 | tx->tx_txg = 0; | |
1071 | } | |
1072 | ||
1073 | /* | |
1074 | * Assign tx to a transaction group. txg_how can be one of: | |
1075 | * | |
1076 | * (1) TXG_WAIT. If the current open txg is full, waits until there's | |
1077 | * a new one. This should be used when you're not holding locks. | |
1078 | * If will only fail if we're truly out of space (or over quota). | |
1079 | * | |
1080 | * (2) TXG_NOWAIT. If we can't assign into the current open txg without | |
1081 | * blocking, returns immediately with ERESTART. This should be used | |
1082 | * whenever you're holding locks. On an ERESTART error, the caller | |
1083 | * should drop locks, do a dmu_tx_wait(tx), and try again. | |
1084 | * | |
1085 | * (3) A specific txg. Use this if you need to ensure that multiple | |
1086 | * transactions all sync in the same txg. Like TXG_NOWAIT, it | |
1087 | * returns ERESTART if it can't assign you into the requested txg. | |
1088 | */ | |
1089 | int | |
1090 | dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how) | |
1091 | { | |
92334b14 | 1092 | hrtime_t before, after; |
34dc7c2f BB |
1093 | int err; |
1094 | ||
1095 | ASSERT(tx->tx_txg == 0); | |
1096 | ASSERT(txg_how != 0); | |
1097 | ASSERT(!dsl_pool_sync_context(tx->tx_pool)); | |
1098 | ||
92334b14 PS |
1099 | before = gethrtime(); |
1100 | ||
34dc7c2f BB |
1101 | while ((err = dmu_tx_try_assign(tx, txg_how)) != 0) { |
1102 | dmu_tx_unassign(tx); | |
1103 | ||
1104 | if (err != ERESTART || txg_how != TXG_WAIT) | |
1105 | return (err); | |
1106 | ||
1107 | dmu_tx_wait(tx); | |
1108 | } | |
1109 | ||
1110 | txg_rele_to_quiesce(&tx->tx_txgh); | |
1111 | ||
92334b14 PS |
1112 | after = gethrtime(); |
1113 | ||
1114 | dsl_pool_tx_assign_add_usecs(tx->tx_pool, | |
1115 | (after - before) / NSEC_PER_USEC); | |
1116 | ||
34dc7c2f BB |
1117 | return (0); |
1118 | } | |
1119 | ||
1120 | void | |
1121 | dmu_tx_wait(dmu_tx_t *tx) | |
1122 | { | |
1123 | spa_t *spa = tx->tx_pool->dp_spa; | |
1124 | ||
1125 | ASSERT(tx->tx_txg == 0); | |
1126 | ||
1127 | /* | |
1128 | * It's possible that the pool has become active after this thread | |
1129 | * has tried to obtain a tx. If that's the case then his | |
1130 | * tx_lasttried_txg would not have been assigned. | |
1131 | */ | |
b128c09f | 1132 | if (spa_suspended(spa) || tx->tx_lasttried_txg == 0) { |
34dc7c2f BB |
1133 | txg_wait_synced(tx->tx_pool, spa_last_synced_txg(spa) + 1); |
1134 | } else if (tx->tx_needassign_txh) { | |
1135 | dnode_t *dn = tx->tx_needassign_txh->txh_dnode; | |
1136 | ||
1137 | mutex_enter(&dn->dn_mtx); | |
1138 | while (dn->dn_assigned_txg == tx->tx_lasttried_txg - 1) | |
1139 | cv_wait(&dn->dn_notxholds, &dn->dn_mtx); | |
1140 | mutex_exit(&dn->dn_mtx); | |
1141 | tx->tx_needassign_txh = NULL; | |
1142 | } else { | |
1143 | txg_wait_open(tx->tx_pool, tx->tx_lasttried_txg + 1); | |
1144 | } | |
1145 | } | |
1146 | ||
1147 | void | |
1148 | dmu_tx_willuse_space(dmu_tx_t *tx, int64_t delta) | |
1149 | { | |
1c5de20a | 1150 | #ifdef DEBUG_DMU_TX |
34dc7c2f BB |
1151 | if (tx->tx_dir == NULL || delta == 0) |
1152 | return; | |
1153 | ||
1154 | if (delta > 0) { | |
1155 | ASSERT3U(refcount_count(&tx->tx_space_written) + delta, <=, | |
1156 | tx->tx_space_towrite); | |
1157 | (void) refcount_add_many(&tx->tx_space_written, delta, NULL); | |
1158 | } else { | |
1159 | (void) refcount_add_many(&tx->tx_space_freed, -delta, NULL); | |
1160 | } | |
1161 | #endif | |
1162 | } | |
1163 | ||
1164 | void | |
1165 | dmu_tx_commit(dmu_tx_t *tx) | |
1166 | { | |
1167 | dmu_tx_hold_t *txh; | |
1168 | ||
1169 | ASSERT(tx->tx_txg != 0); | |
1170 | ||
c65aa5b2 | 1171 | while ((txh = list_head(&tx->tx_holds))) { |
34dc7c2f BB |
1172 | dnode_t *dn = txh->txh_dnode; |
1173 | ||
1174 | list_remove(&tx->tx_holds, txh); | |
1175 | kmem_free(txh, sizeof (dmu_tx_hold_t)); | |
1176 | if (dn == NULL) | |
1177 | continue; | |
1178 | mutex_enter(&dn->dn_mtx); | |
1179 | ASSERT3U(dn->dn_assigned_txg, ==, tx->tx_txg); | |
1180 | ||
1181 | if (refcount_remove(&dn->dn_tx_holds, tx) == 0) { | |
1182 | dn->dn_assigned_txg = 0; | |
1183 | cv_broadcast(&dn->dn_notxholds); | |
1184 | } | |
1185 | mutex_exit(&dn->dn_mtx); | |
1186 | dnode_rele(dn, tx); | |
1187 | } | |
1188 | ||
1189 | if (tx->tx_tempreserve_cookie) | |
1190 | dsl_dir_tempreserve_clear(tx->tx_tempreserve_cookie, tx); | |
1191 | ||
428870ff BB |
1192 | if (!list_is_empty(&tx->tx_callbacks)) |
1193 | txg_register_callbacks(&tx->tx_txgh, &tx->tx_callbacks); | |
1194 | ||
34dc7c2f BB |
1195 | if (tx->tx_anyobj == FALSE) |
1196 | txg_rele_to_sync(&tx->tx_txgh); | |
428870ff BB |
1197 | |
1198 | list_destroy(&tx->tx_callbacks); | |
34dc7c2f | 1199 | list_destroy(&tx->tx_holds); |
1c5de20a | 1200 | #ifdef DEBUG_DMU_TX |
34dc7c2f BB |
1201 | dprintf("towrite=%llu written=%llu tofree=%llu freed=%llu\n", |
1202 | tx->tx_space_towrite, refcount_count(&tx->tx_space_written), | |
1203 | tx->tx_space_tofree, refcount_count(&tx->tx_space_freed)); | |
1204 | refcount_destroy_many(&tx->tx_space_written, | |
1205 | refcount_count(&tx->tx_space_written)); | |
1206 | refcount_destroy_many(&tx->tx_space_freed, | |
1207 | refcount_count(&tx->tx_space_freed)); | |
1208 | #endif | |
1209 | kmem_free(tx, sizeof (dmu_tx_t)); | |
1210 | } | |
1211 | ||
1212 | void | |
1213 | dmu_tx_abort(dmu_tx_t *tx) | |
1214 | { | |
1215 | dmu_tx_hold_t *txh; | |
1216 | ||
1217 | ASSERT(tx->tx_txg == 0); | |
1218 | ||
c65aa5b2 | 1219 | while ((txh = list_head(&tx->tx_holds))) { |
34dc7c2f BB |
1220 | dnode_t *dn = txh->txh_dnode; |
1221 | ||
1222 | list_remove(&tx->tx_holds, txh); | |
1223 | kmem_free(txh, sizeof (dmu_tx_hold_t)); | |
1224 | if (dn != NULL) | |
1225 | dnode_rele(dn, tx); | |
1226 | } | |
428870ff BB |
1227 | |
1228 | /* | |
1229 | * Call any registered callbacks with an error code. | |
1230 | */ | |
1231 | if (!list_is_empty(&tx->tx_callbacks)) | |
1232 | dmu_tx_do_callbacks(&tx->tx_callbacks, ECANCELED); | |
1233 | ||
1234 | list_destroy(&tx->tx_callbacks); | |
34dc7c2f | 1235 | list_destroy(&tx->tx_holds); |
1c5de20a | 1236 | #ifdef DEBUG_DMU_TX |
34dc7c2f BB |
1237 | refcount_destroy_many(&tx->tx_space_written, |
1238 | refcount_count(&tx->tx_space_written)); | |
1239 | refcount_destroy_many(&tx->tx_space_freed, | |
1240 | refcount_count(&tx->tx_space_freed)); | |
1241 | #endif | |
1242 | kmem_free(tx, sizeof (dmu_tx_t)); | |
1243 | } | |
1244 | ||
1245 | uint64_t | |
1246 | dmu_tx_get_txg(dmu_tx_t *tx) | |
1247 | { | |
1248 | ASSERT(tx->tx_txg != 0); | |
1249 | return (tx->tx_txg); | |
1250 | } | |
428870ff BB |
1251 | |
1252 | void | |
1253 | dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *func, void *data) | |
1254 | { | |
1255 | dmu_tx_callback_t *dcb; | |
1256 | ||
b8d06fca | 1257 | dcb = kmem_alloc(sizeof (dmu_tx_callback_t), KM_PUSHPAGE); |
428870ff BB |
1258 | |
1259 | dcb->dcb_func = func; | |
1260 | dcb->dcb_data = data; | |
1261 | ||
1262 | list_insert_tail(&tx->tx_callbacks, dcb); | |
1263 | } | |
1264 | ||
1265 | /* | |
1266 | * Call all the commit callbacks on a list, with a given error code. | |
1267 | */ | |
1268 | void | |
1269 | dmu_tx_do_callbacks(list_t *cb_list, int error) | |
1270 | { | |
1271 | dmu_tx_callback_t *dcb; | |
1272 | ||
c65aa5b2 | 1273 | while ((dcb = list_head(cb_list))) { |
428870ff BB |
1274 | list_remove(cb_list, dcb); |
1275 | dcb->dcb_func(dcb->dcb_data, error); | |
1276 | kmem_free(dcb, sizeof (dmu_tx_callback_t)); | |
1277 | } | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * Interface to hold a bunch of attributes. | |
1282 | * used for creating new files. | |
1283 | * attrsize is the total size of all attributes | |
1284 | * to be added during object creation | |
1285 | * | |
1286 | * For updating/adding a single attribute dmu_tx_hold_sa() should be used. | |
1287 | */ | |
1288 | ||
1289 | /* | |
1290 | * hold necessary attribute name for attribute registration. | |
1291 | * should be a very rare case where this is needed. If it does | |
1292 | * happen it would only happen on the first write to the file system. | |
1293 | */ | |
1294 | static void | |
1295 | dmu_tx_sa_registration_hold(sa_os_t *sa, dmu_tx_t *tx) | |
1296 | { | |
1297 | int i; | |
1298 | ||
1299 | if (!sa->sa_need_attr_registration) | |
1300 | return; | |
1301 | ||
1302 | for (i = 0; i != sa->sa_num_attrs; i++) { | |
1303 | if (!sa->sa_attr_table[i].sa_registered) { | |
1304 | if (sa->sa_reg_attr_obj) | |
1305 | dmu_tx_hold_zap(tx, sa->sa_reg_attr_obj, | |
1306 | B_TRUE, sa->sa_attr_table[i].sa_name); | |
1307 | else | |
1308 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, | |
1309 | B_TRUE, sa->sa_attr_table[i].sa_name); | |
1310 | } | |
1311 | } | |
1312 | } | |
1313 | ||
1314 | ||
1315 | void | |
1316 | dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object) | |
1317 | { | |
1318 | dnode_t *dn; | |
1319 | dmu_tx_hold_t *txh; | |
428870ff BB |
1320 | |
1321 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, object, | |
1322 | THT_SPILL, 0, 0); | |
1323 | ||
1324 | dn = txh->txh_dnode; | |
1325 | ||
1326 | if (dn == NULL) | |
1327 | return; | |
1328 | ||
1329 | /* If blkptr doesn't exist then add space to towrite */ | |
22cd4a46 | 1330 | if (!(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) { |
428870ff | 1331 | txh->txh_space_towrite += SPA_MAXBLOCKSIZE; |
428870ff | 1332 | } else { |
22cd4a46 AL |
1333 | blkptr_t *bp; |
1334 | ||
1335 | bp = &dn->dn_phys->dn_spill; | |
428870ff BB |
1336 | if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset, |
1337 | bp, bp->blk_birth)) | |
1338 | txh->txh_space_tooverwrite += SPA_MAXBLOCKSIZE; | |
1339 | else | |
1340 | txh->txh_space_towrite += SPA_MAXBLOCKSIZE; | |
22cd4a46 | 1341 | if (!BP_IS_HOLE(bp)) |
428870ff BB |
1342 | txh->txh_space_tounref += SPA_MAXBLOCKSIZE; |
1343 | } | |
1344 | } | |
1345 | ||
1346 | void | |
1347 | dmu_tx_hold_sa_create(dmu_tx_t *tx, int attrsize) | |
1348 | { | |
1349 | sa_os_t *sa = tx->tx_objset->os_sa; | |
1350 | ||
1351 | dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); | |
1352 | ||
1353 | if (tx->tx_objset->os_sa->sa_master_obj == 0) | |
1354 | return; | |
1355 | ||
1356 | if (tx->tx_objset->os_sa->sa_layout_attr_obj) | |
1357 | dmu_tx_hold_zap(tx, sa->sa_layout_attr_obj, B_TRUE, NULL); | |
1358 | else { | |
1359 | dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_LAYOUTS); | |
1360 | dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_REGISTRY); | |
1361 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1362 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1363 | } | |
1364 | ||
1365 | dmu_tx_sa_registration_hold(sa, tx); | |
1366 | ||
1367 | if (attrsize <= DN_MAX_BONUSLEN && !sa->sa_force_spill) | |
1368 | return; | |
1369 | ||
1370 | (void) dmu_tx_hold_object_impl(tx, tx->tx_objset, DMU_NEW_OBJECT, | |
1371 | THT_SPILL, 0, 0); | |
1372 | } | |
1373 | ||
1374 | /* | |
1375 | * Hold SA attribute | |
1376 | * | |
1377 | * dmu_tx_hold_sa(dmu_tx_t *tx, sa_handle_t *, attribute, add, size) | |
1378 | * | |
1379 | * variable_size is the total size of all variable sized attributes | |
1380 | * passed to this function. It is not the total size of all | |
1381 | * variable size attributes that *may* exist on this object. | |
1382 | */ | |
1383 | void | |
1384 | dmu_tx_hold_sa(dmu_tx_t *tx, sa_handle_t *hdl, boolean_t may_grow) | |
1385 | { | |
1386 | uint64_t object; | |
1387 | sa_os_t *sa = tx->tx_objset->os_sa; | |
1388 | ||
1389 | ASSERT(hdl != NULL); | |
1390 | ||
1391 | object = sa_handle_object(hdl); | |
1392 | ||
1393 | dmu_tx_hold_bonus(tx, object); | |
1394 | ||
1395 | if (tx->tx_objset->os_sa->sa_master_obj == 0) | |
1396 | return; | |
1397 | ||
1398 | if (tx->tx_objset->os_sa->sa_reg_attr_obj == 0 || | |
1399 | tx->tx_objset->os_sa->sa_layout_attr_obj == 0) { | |
1400 | dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_LAYOUTS); | |
1401 | dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_REGISTRY); | |
1402 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1403 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1404 | } | |
1405 | ||
1406 | dmu_tx_sa_registration_hold(sa, tx); | |
1407 | ||
1408 | if (may_grow && tx->tx_objset->os_sa->sa_layout_attr_obj) | |
1409 | dmu_tx_hold_zap(tx, sa->sa_layout_attr_obj, B_TRUE, NULL); | |
1410 | ||
572e2857 | 1411 | if (sa->sa_force_spill || may_grow || hdl->sa_spill) { |
428870ff BB |
1412 | ASSERT(tx->tx_txg == 0); |
1413 | dmu_tx_hold_spill(tx, object); | |
572e2857 BB |
1414 | } else { |
1415 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)hdl->sa_bonus; | |
1416 | dnode_t *dn; | |
1417 | ||
1418 | DB_DNODE_ENTER(db); | |
1419 | dn = DB_DNODE(db); | |
1420 | if (dn->dn_have_spill) { | |
1421 | ASSERT(tx->tx_txg == 0); | |
1422 | dmu_tx_hold_spill(tx, object); | |
1423 | } | |
1424 | DB_DNODE_EXIT(db); | |
428870ff BB |
1425 | } |
1426 | } | |
c28b2279 | 1427 | |
570827e1 BB |
1428 | void |
1429 | dmu_tx_init(void) | |
1430 | { | |
1431 | dmu_tx_ksp = kstat_create("zfs", 0, "dmu_tx", "misc", | |
1432 | KSTAT_TYPE_NAMED, sizeof (dmu_tx_stats) / sizeof (kstat_named_t), | |
1433 | KSTAT_FLAG_VIRTUAL); | |
1434 | ||
1435 | if (dmu_tx_ksp != NULL) { | |
1436 | dmu_tx_ksp->ks_data = &dmu_tx_stats; | |
1437 | kstat_install(dmu_tx_ksp); | |
1438 | } | |
1439 | } | |
1440 | ||
1441 | void | |
1442 | dmu_tx_fini(void) | |
1443 | { | |
1444 | if (dmu_tx_ksp != NULL) { | |
1445 | kstat_delete(dmu_tx_ksp); | |
1446 | dmu_tx_ksp = NULL; | |
1447 | } | |
1448 | } | |
1449 | ||
c28b2279 BB |
1450 | #if defined(_KERNEL) && defined(HAVE_SPL) |
1451 | EXPORT_SYMBOL(dmu_tx_create); | |
1452 | EXPORT_SYMBOL(dmu_tx_hold_write); | |
1453 | EXPORT_SYMBOL(dmu_tx_hold_free); | |
1454 | EXPORT_SYMBOL(dmu_tx_hold_zap); | |
1455 | EXPORT_SYMBOL(dmu_tx_hold_bonus); | |
1456 | EXPORT_SYMBOL(dmu_tx_abort); | |
1457 | EXPORT_SYMBOL(dmu_tx_assign); | |
1458 | EXPORT_SYMBOL(dmu_tx_wait); | |
1459 | EXPORT_SYMBOL(dmu_tx_commit); | |
1460 | EXPORT_SYMBOL(dmu_tx_get_txg); | |
1461 | EXPORT_SYMBOL(dmu_tx_callback_register); | |
1462 | EXPORT_SYMBOL(dmu_tx_do_callbacks); | |
1463 | EXPORT_SYMBOL(dmu_tx_hold_spill); | |
1464 | EXPORT_SYMBOL(dmu_tx_hold_sa_create); | |
1465 | EXPORT_SYMBOL(dmu_tx_hold_sa); | |
1466 | #endif |