]>
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. |
34dc7c2f BB |
23 | */ |
24 | ||
34dc7c2f BB |
25 | #include <sys/dmu.h> |
26 | #include <sys/dmu_impl.h> | |
27 | #include <sys/dbuf.h> | |
28 | #include <sys/dmu_tx.h> | |
29 | #include <sys/dmu_objset.h> | |
30 | #include <sys/dsl_dataset.h> /* for dsl_dataset_block_freeable() */ | |
31 | #include <sys/dsl_dir.h> /* for dsl_dir_tempreserve_*() */ | |
32 | #include <sys/dsl_pool.h> | |
33 | #include <sys/zap_impl.h> /* for fzap_default_block_shift */ | |
34 | #include <sys/spa.h> | |
428870ff BB |
35 | #include <sys/sa.h> |
36 | #include <sys/sa_impl.h> | |
34dc7c2f | 37 | #include <sys/zfs_context.h> |
428870ff | 38 | #include <sys/varargs.h> |
34dc7c2f BB |
39 | |
40 | typedef void (*dmu_tx_hold_func_t)(dmu_tx_t *tx, struct dnode *dn, | |
41 | uint64_t arg1, uint64_t arg2); | |
42 | ||
43 | ||
44 | dmu_tx_t * | |
45 | dmu_tx_create_dd(dsl_dir_t *dd) | |
46 | { | |
47 | dmu_tx_t *tx = kmem_zalloc(sizeof (dmu_tx_t), KM_SLEEP); | |
48 | tx->tx_dir = dd; | |
49 | if (dd) | |
50 | tx->tx_pool = dd->dd_pool; | |
51 | list_create(&tx->tx_holds, sizeof (dmu_tx_hold_t), | |
52 | offsetof(dmu_tx_hold_t, txh_node)); | |
428870ff BB |
53 | list_create(&tx->tx_callbacks, sizeof (dmu_tx_callback_t), |
54 | offsetof(dmu_tx_callback_t, dcb_node)); | |
34dc7c2f BB |
55 | #ifdef ZFS_DEBUG |
56 | refcount_create(&tx->tx_space_written); | |
57 | refcount_create(&tx->tx_space_freed); | |
58 | #endif | |
59 | return (tx); | |
60 | } | |
61 | ||
62 | dmu_tx_t * | |
63 | dmu_tx_create(objset_t *os) | |
64 | { | |
428870ff | 65 | dmu_tx_t *tx = dmu_tx_create_dd(os->os_dsl_dataset->ds_dir); |
34dc7c2f | 66 | tx->tx_objset = os; |
428870ff | 67 | tx->tx_lastsnap_txg = dsl_dataset_prev_snap_txg(os->os_dsl_dataset); |
34dc7c2f BB |
68 | return (tx); |
69 | } | |
70 | ||
71 | dmu_tx_t * | |
72 | dmu_tx_create_assigned(struct dsl_pool *dp, uint64_t txg) | |
73 | { | |
74 | dmu_tx_t *tx = dmu_tx_create_dd(NULL); | |
75 | ||
76 | ASSERT3U(txg, <=, dp->dp_tx.tx_open_txg); | |
77 | tx->tx_pool = dp; | |
78 | tx->tx_txg = txg; | |
79 | tx->tx_anyobj = TRUE; | |
80 | ||
81 | return (tx); | |
82 | } | |
83 | ||
84 | int | |
85 | dmu_tx_is_syncing(dmu_tx_t *tx) | |
86 | { | |
87 | return (tx->tx_anyobj); | |
88 | } | |
89 | ||
90 | int | |
91 | dmu_tx_private_ok(dmu_tx_t *tx) | |
92 | { | |
93 | return (tx->tx_anyobj); | |
94 | } | |
95 | ||
96 | static dmu_tx_hold_t * | |
97 | dmu_tx_hold_object_impl(dmu_tx_t *tx, objset_t *os, uint64_t object, | |
98 | enum dmu_tx_hold_type type, uint64_t arg1, uint64_t arg2) | |
99 | { | |
100 | dmu_tx_hold_t *txh; | |
101 | dnode_t *dn = NULL; | |
102 | int err; | |
103 | ||
104 | if (object != DMU_NEW_OBJECT) { | |
428870ff | 105 | err = dnode_hold(os, object, tx, &dn); |
34dc7c2f BB |
106 | if (err) { |
107 | tx->tx_err = err; | |
108 | return (NULL); | |
109 | } | |
110 | ||
111 | if (err == 0 && tx->tx_txg != 0) { | |
112 | mutex_enter(&dn->dn_mtx); | |
113 | /* | |
114 | * dn->dn_assigned_txg == tx->tx_txg doesn't pose a | |
115 | * problem, but there's no way for it to happen (for | |
116 | * now, at least). | |
117 | */ | |
118 | ASSERT(dn->dn_assigned_txg == 0); | |
119 | dn->dn_assigned_txg = tx->tx_txg; | |
120 | (void) refcount_add(&dn->dn_tx_holds, tx); | |
121 | mutex_exit(&dn->dn_mtx); | |
122 | } | |
123 | } | |
124 | ||
125 | txh = kmem_zalloc(sizeof (dmu_tx_hold_t), KM_SLEEP); | |
126 | txh->txh_tx = tx; | |
127 | txh->txh_dnode = dn; | |
128 | #ifdef ZFS_DEBUG | |
129 | txh->txh_type = type; | |
130 | txh->txh_arg1 = arg1; | |
131 | txh->txh_arg2 = arg2; | |
132 | #endif | |
133 | list_insert_tail(&tx->tx_holds, txh); | |
134 | ||
135 | return (txh); | |
136 | } | |
137 | ||
138 | void | |
139 | dmu_tx_add_new_object(dmu_tx_t *tx, objset_t *os, uint64_t object) | |
140 | { | |
141 | /* | |
142 | * If we're syncing, they can manipulate any object anyhow, and | |
143 | * the hold on the dnode_t can cause problems. | |
144 | */ | |
145 | if (!dmu_tx_is_syncing(tx)) { | |
146 | (void) dmu_tx_hold_object_impl(tx, os, | |
147 | object, THT_NEWOBJECT, 0, 0); | |
148 | } | |
149 | } | |
150 | ||
151 | static int | |
152 | dmu_tx_check_ioerr(zio_t *zio, dnode_t *dn, int level, uint64_t blkid) | |
153 | { | |
154 | int err; | |
155 | dmu_buf_impl_t *db; | |
156 | ||
157 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
158 | db = dbuf_hold_level(dn, level, blkid, FTAG); | |
159 | rw_exit(&dn->dn_struct_rwlock); | |
160 | if (db == NULL) | |
161 | return (EIO); | |
162 | err = dbuf_read(db, zio, DB_RF_CANFAIL | DB_RF_NOPREFETCH); | |
163 | dbuf_rele(db, FTAG); | |
164 | return (err); | |
165 | } | |
166 | ||
9babb374 | 167 | static void |
428870ff BB |
168 | dmu_tx_count_twig(dmu_tx_hold_t *txh, dnode_t *dn, dmu_buf_impl_t *db, |
169 | int level, uint64_t blkid, boolean_t freeable, uint64_t *history) | |
9babb374 | 170 | { |
428870ff BB |
171 | objset_t *os = dn->dn_objset; |
172 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
173 | int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
174 | dmu_buf_impl_t *parent = NULL; | |
175 | blkptr_t *bp = NULL; | |
176 | uint64_t space; | |
177 | ||
178 | if (level >= dn->dn_nlevels || history[level] == blkid) | |
9babb374 BB |
179 | return; |
180 | ||
428870ff | 181 | history[level] = blkid; |
9babb374 | 182 | |
428870ff BB |
183 | space = (level == 0) ? dn->dn_datablksz : (1ULL << dn->dn_indblkshift); |
184 | ||
185 | if (db == NULL || db == dn->dn_dbuf) { | |
186 | ASSERT(level != 0); | |
187 | db = NULL; | |
188 | } else { | |
189 | ASSERT(db->db_dnode == dn); | |
190 | ASSERT(db->db_level == level); | |
191 | ASSERT(db->db.db_size == space); | |
192 | ASSERT(db->db_blkid == blkid); | |
193 | bp = db->db_blkptr; | |
194 | parent = db->db_parent; | |
9babb374 BB |
195 | } |
196 | ||
428870ff BB |
197 | freeable = (bp && (freeable || |
198 | dsl_dataset_block_freeable(ds, bp, bp->blk_birth))); | |
9babb374 | 199 | |
428870ff BB |
200 | if (freeable) |
201 | txh->txh_space_tooverwrite += space; | |
202 | else | |
203 | txh->txh_space_towrite += space; | |
204 | if (bp) | |
205 | txh->txh_space_tounref += bp_get_dsize(os->os_spa, bp); | |
206 | ||
207 | dmu_tx_count_twig(txh, dn, parent, level + 1, | |
208 | blkid >> epbs, freeable, history); | |
9babb374 BB |
209 | } |
210 | ||
34dc7c2f BB |
211 | /* ARGSUSED */ |
212 | static void | |
213 | dmu_tx_count_write(dmu_tx_hold_t *txh, uint64_t off, uint64_t len) | |
214 | { | |
215 | dnode_t *dn = txh->txh_dnode; | |
216 | uint64_t start, end, i; | |
217 | int min_bs, max_bs, min_ibs, max_ibs, epbs, bits; | |
218 | int err = 0; | |
219 | ||
220 | if (len == 0) | |
221 | return; | |
222 | ||
223 | min_bs = SPA_MINBLOCKSHIFT; | |
224 | max_bs = SPA_MAXBLOCKSHIFT; | |
225 | min_ibs = DN_MIN_INDBLKSHIFT; | |
226 | max_ibs = DN_MAX_INDBLKSHIFT; | |
227 | ||
34dc7c2f | 228 | if (dn) { |
428870ff | 229 | uint64_t history[DN_MAX_LEVELS]; |
9babb374 BB |
230 | int nlvls = dn->dn_nlevels; |
231 | int delta; | |
232 | ||
233 | /* | |
234 | * For i/o error checking, read the first and last level-0 | |
235 | * blocks (if they are not aligned), and all the level-1 blocks. | |
236 | */ | |
34dc7c2f | 237 | if (dn->dn_maxblkid == 0) { |
9babb374 BB |
238 | delta = dn->dn_datablksz; |
239 | start = (off < dn->dn_datablksz) ? 0 : 1; | |
240 | end = (off+len <= dn->dn_datablksz) ? 0 : 1; | |
241 | if (start == 0 && (off > 0 || len < dn->dn_datablksz)) { | |
b128c09f BB |
242 | err = dmu_tx_check_ioerr(NULL, dn, 0, 0); |
243 | if (err) | |
244 | goto out; | |
9babb374 | 245 | delta -= off; |
b128c09f | 246 | } |
34dc7c2f BB |
247 | } else { |
248 | zio_t *zio = zio_root(dn->dn_objset->os_spa, | |
249 | NULL, NULL, ZIO_FLAG_CANFAIL); | |
250 | ||
251 | /* first level-0 block */ | |
252 | start = off >> dn->dn_datablkshift; | |
253 | if (P2PHASE(off, dn->dn_datablksz) || | |
254 | len < dn->dn_datablksz) { | |
255 | err = dmu_tx_check_ioerr(zio, dn, 0, start); | |
256 | if (err) | |
257 | goto out; | |
258 | } | |
259 | ||
260 | /* last level-0 block */ | |
261 | end = (off+len-1) >> dn->dn_datablkshift; | |
b128c09f | 262 | if (end != start && end <= dn->dn_maxblkid && |
34dc7c2f BB |
263 | P2PHASE(off+len, dn->dn_datablksz)) { |
264 | err = dmu_tx_check_ioerr(zio, dn, 0, end); | |
265 | if (err) | |
266 | goto out; | |
267 | } | |
268 | ||
269 | /* level-1 blocks */ | |
9babb374 BB |
270 | if (nlvls > 1) { |
271 | int shft = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
272 | for (i = (start>>shft)+1; i < end>>shft; i++) { | |
34dc7c2f BB |
273 | err = dmu_tx_check_ioerr(zio, dn, 1, i); |
274 | if (err) | |
275 | goto out; | |
276 | } | |
277 | } | |
278 | ||
279 | err = zio_wait(zio); | |
280 | if (err) | |
281 | goto out; | |
9babb374 | 282 | delta = P2NPHASE(off, dn->dn_datablksz); |
34dc7c2f | 283 | } |
34dc7c2f | 284 | |
9babb374 BB |
285 | if (dn->dn_maxblkid > 0) { |
286 | /* | |
287 | * The blocksize can't change, | |
288 | * so we can make a more precise estimate. | |
289 | */ | |
290 | ASSERT(dn->dn_datablkshift != 0); | |
34dc7c2f | 291 | min_bs = max_bs = dn->dn_datablkshift; |
9babb374 BB |
292 | min_ibs = max_ibs = dn->dn_indblkshift; |
293 | } else if (dn->dn_indblkshift > max_ibs) { | |
294 | /* | |
295 | * This ensures that if we reduce DN_MAX_INDBLKSHIFT, | |
296 | * the code will still work correctly on older pools. | |
297 | */ | |
298 | min_ibs = max_ibs = dn->dn_indblkshift; | |
299 | } | |
300 | ||
301 | /* | |
302 | * If this write is not off the end of the file | |
303 | * we need to account for overwrites/unref. | |
304 | */ | |
428870ff BB |
305 | if (start <= dn->dn_maxblkid) { |
306 | for (int l = 0; l < DN_MAX_LEVELS; l++) | |
307 | history[l] = -1ULL; | |
308 | } | |
9babb374 | 309 | while (start <= dn->dn_maxblkid) { |
9babb374 BB |
310 | dmu_buf_impl_t *db; |
311 | ||
312 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
428870ff | 313 | err = dbuf_hold_impl(dn, 0, start, FALSE, FTAG, &db); |
9babb374 | 314 | rw_exit(&dn->dn_struct_rwlock); |
428870ff BB |
315 | |
316 | if (err) { | |
317 | txh->txh_tx->tx_err = err; | |
318 | return; | |
9babb374 | 319 | } |
428870ff BB |
320 | |
321 | dmu_tx_count_twig(txh, dn, db, 0, start, B_FALSE, | |
322 | history); | |
9babb374 BB |
323 | dbuf_rele(db, FTAG); |
324 | if (++start > end) { | |
325 | /* | |
326 | * Account for new indirects appearing | |
327 | * before this IO gets assigned into a txg. | |
328 | */ | |
329 | bits = 64 - min_bs; | |
330 | epbs = min_ibs - SPA_BLKPTRSHIFT; | |
331 | for (bits -= epbs * (nlvls - 1); | |
332 | bits >= 0; bits -= epbs) | |
333 | txh->txh_fudge += 1ULL << max_ibs; | |
334 | goto out; | |
335 | } | |
336 | off += delta; | |
337 | if (len >= delta) | |
338 | len -= delta; | |
339 | delta = dn->dn_datablksz; | |
340 | } | |
34dc7c2f BB |
341 | } |
342 | ||
343 | /* | |
344 | * 'end' is the last thing we will access, not one past. | |
345 | * This way we won't overflow when accessing the last byte. | |
346 | */ | |
347 | start = P2ALIGN(off, 1ULL << max_bs); | |
348 | end = P2ROUNDUP(off + len, 1ULL << max_bs) - 1; | |
349 | txh->txh_space_towrite += end - start + 1; | |
350 | ||
351 | start >>= min_bs; | |
352 | end >>= min_bs; | |
353 | ||
354 | epbs = min_ibs - SPA_BLKPTRSHIFT; | |
355 | ||
356 | /* | |
357 | * The object contains at most 2^(64 - min_bs) blocks, | |
358 | * and each indirect level maps 2^epbs. | |
359 | */ | |
360 | for (bits = 64 - min_bs; bits >= 0; bits -= epbs) { | |
361 | start >>= epbs; | |
362 | end >>= epbs; | |
9babb374 | 363 | ASSERT3U(end, >=, start); |
34dc7c2f | 364 | txh->txh_space_towrite += (end - start + 1) << max_ibs; |
9babb374 BB |
365 | if (start != 0) { |
366 | /* | |
367 | * We also need a new blkid=0 indirect block | |
368 | * to reference any existing file data. | |
369 | */ | |
370 | txh->txh_space_towrite += 1ULL << max_ibs; | |
371 | } | |
34dc7c2f BB |
372 | } |
373 | ||
34dc7c2f | 374 | out: |
9babb374 BB |
375 | if (txh->txh_space_towrite + txh->txh_space_tooverwrite > |
376 | 2 * DMU_MAX_ACCESS) | |
377 | err = EFBIG; | |
378 | ||
34dc7c2f BB |
379 | if (err) |
380 | txh->txh_tx->tx_err = err; | |
381 | } | |
382 | ||
383 | static void | |
384 | dmu_tx_count_dnode(dmu_tx_hold_t *txh) | |
385 | { | |
386 | dnode_t *dn = txh->txh_dnode; | |
428870ff | 387 | dnode_t *mdn = txh->txh_tx->tx_objset->os_meta_dnode; |
34dc7c2f BB |
388 | uint64_t space = mdn->dn_datablksz + |
389 | ((mdn->dn_nlevels-1) << mdn->dn_indblkshift); | |
390 | ||
391 | if (dn && dn->dn_dbuf->db_blkptr && | |
392 | dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset, | |
428870ff | 393 | dn->dn_dbuf->db_blkptr, dn->dn_dbuf->db_blkptr->blk_birth)) { |
34dc7c2f | 394 | txh->txh_space_tooverwrite += space; |
9babb374 | 395 | txh->txh_space_tounref += space; |
34dc7c2f BB |
396 | } else { |
397 | txh->txh_space_towrite += space; | |
398 | if (dn && dn->dn_dbuf->db_blkptr) | |
399 | txh->txh_space_tounref += space; | |
400 | } | |
401 | } | |
402 | ||
403 | void | |
404 | dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len) | |
405 | { | |
406 | dmu_tx_hold_t *txh; | |
407 | ||
408 | ASSERT(tx->tx_txg == 0); | |
409 | ASSERT(len < DMU_MAX_ACCESS); | |
410 | ASSERT(len == 0 || UINT64_MAX - off >= len - 1); | |
411 | ||
412 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, | |
413 | object, THT_WRITE, off, len); | |
414 | if (txh == NULL) | |
415 | return; | |
416 | ||
417 | dmu_tx_count_write(txh, off, len); | |
418 | dmu_tx_count_dnode(txh); | |
419 | } | |
420 | ||
421 | static void | |
422 | dmu_tx_count_free(dmu_tx_hold_t *txh, uint64_t off, uint64_t len) | |
423 | { | |
b128c09f BB |
424 | uint64_t blkid, nblks, lastblk; |
425 | uint64_t space = 0, unref = 0, skipped = 0; | |
34dc7c2f BB |
426 | dnode_t *dn = txh->txh_dnode; |
427 | dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; | |
428 | spa_t *spa = txh->txh_tx->tx_pool->dp_spa; | |
b128c09f | 429 | int epbs; |
34dc7c2f | 430 | |
b128c09f | 431 | if (dn->dn_nlevels == 0) |
34dc7c2f BB |
432 | return; |
433 | ||
434 | /* | |
b128c09f | 435 | * The struct_rwlock protects us against dn_nlevels |
34dc7c2f BB |
436 | * changing, in case (against all odds) we manage to dirty & |
437 | * sync out the changes after we check for being dirty. | |
428870ff | 438 | * Also, dbuf_hold_impl() wants us to have the struct_rwlock. |
34dc7c2f BB |
439 | */ |
440 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
b128c09f BB |
441 | epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; |
442 | if (dn->dn_maxblkid == 0) { | |
34dc7c2f BB |
443 | if (off == 0 && len >= dn->dn_datablksz) { |
444 | blkid = 0; | |
445 | nblks = 1; | |
446 | } else { | |
447 | rw_exit(&dn->dn_struct_rwlock); | |
448 | return; | |
449 | } | |
450 | } else { | |
451 | blkid = off >> dn->dn_datablkshift; | |
b128c09f | 452 | nblks = (len + dn->dn_datablksz - 1) >> dn->dn_datablkshift; |
34dc7c2f | 453 | |
b128c09f | 454 | if (blkid >= dn->dn_maxblkid) { |
34dc7c2f BB |
455 | rw_exit(&dn->dn_struct_rwlock); |
456 | return; | |
457 | } | |
b128c09f BB |
458 | if (blkid + nblks > dn->dn_maxblkid) |
459 | nblks = dn->dn_maxblkid - blkid; | |
34dc7c2f | 460 | |
34dc7c2f | 461 | } |
b128c09f | 462 | if (dn->dn_nlevels == 1) { |
34dc7c2f BB |
463 | int i; |
464 | for (i = 0; i < nblks; i++) { | |
465 | blkptr_t *bp = dn->dn_phys->dn_blkptr; | |
b128c09f | 466 | ASSERT3U(blkid + i, <, dn->dn_nblkptr); |
34dc7c2f | 467 | bp += blkid + i; |
428870ff | 468 | if (dsl_dataset_block_freeable(ds, bp, bp->blk_birth)) { |
34dc7c2f | 469 | dprintf_bp(bp, "can free old%s", ""); |
428870ff | 470 | space += bp_get_dsize(spa, bp); |
34dc7c2f BB |
471 | } |
472 | unref += BP_GET_ASIZE(bp); | |
473 | } | |
474 | nblks = 0; | |
475 | } | |
476 | ||
b128c09f BB |
477 | /* |
478 | * Add in memory requirements of higher-level indirects. | |
479 | * This assumes a worst-possible scenario for dn_nlevels. | |
480 | */ | |
481 | { | |
482 | uint64_t blkcnt = 1 + ((nblks >> epbs) >> epbs); | |
483 | int level = (dn->dn_nlevels > 1) ? 2 : 1; | |
484 | ||
485 | while (level++ < DN_MAX_LEVELS) { | |
486 | txh->txh_memory_tohold += blkcnt << dn->dn_indblkshift; | |
487 | blkcnt = 1 + (blkcnt >> epbs); | |
488 | } | |
489 | ASSERT(blkcnt <= dn->dn_nblkptr); | |
490 | } | |
491 | ||
492 | lastblk = blkid + nblks - 1; | |
34dc7c2f BB |
493 | while (nblks) { |
494 | dmu_buf_impl_t *dbuf; | |
b128c09f BB |
495 | uint64_t ibyte, new_blkid; |
496 | int epb = 1 << epbs; | |
497 | int err, i, blkoff, tochk; | |
498 | blkptr_t *bp; | |
499 | ||
500 | ibyte = blkid << dn->dn_datablkshift; | |
501 | err = dnode_next_offset(dn, | |
502 | DNODE_FIND_HAVELOCK, &ibyte, 2, 1, 0); | |
503 | new_blkid = ibyte >> dn->dn_datablkshift; | |
504 | if (err == ESRCH) { | |
505 | skipped += (lastblk >> epbs) - (blkid >> epbs) + 1; | |
506 | break; | |
507 | } | |
508 | if (err) { | |
509 | txh->txh_tx->tx_err = err; | |
510 | break; | |
511 | } | |
512 | if (new_blkid > lastblk) { | |
513 | skipped += (lastblk >> epbs) - (blkid >> epbs) + 1; | |
514 | break; | |
515 | } | |
34dc7c2f | 516 | |
b128c09f BB |
517 | if (new_blkid > blkid) { |
518 | ASSERT((new_blkid >> epbs) > (blkid >> epbs)); | |
519 | skipped += (new_blkid >> epbs) - (blkid >> epbs) - 1; | |
520 | nblks -= new_blkid - blkid; | |
521 | blkid = new_blkid; | |
522 | } | |
523 | blkoff = P2PHASE(blkid, epb); | |
524 | tochk = MIN(epb - blkoff, nblks); | |
34dc7c2f | 525 | |
428870ff BB |
526 | err = dbuf_hold_impl(dn, 1, blkid >> epbs, FALSE, FTAG, &dbuf); |
527 | if (err) { | |
528 | txh->txh_tx->tx_err = err; | |
b128c09f | 529 | break; |
34dc7c2f | 530 | } |
428870ff BB |
531 | |
532 | txh->txh_memory_tohold += dbuf->db.db_size; | |
533 | ||
534 | /* | |
535 | * We don't check memory_tohold against DMU_MAX_ACCESS because | |
536 | * memory_tohold is an over-estimation (especially the >L1 | |
537 | * indirect blocks), so it could fail. Callers should have | |
538 | * already verified that they will not be holding too much | |
539 | * memory. | |
540 | */ | |
541 | ||
b128c09f BB |
542 | err = dbuf_read(dbuf, NULL, DB_RF_HAVESTRUCT | DB_RF_CANFAIL); |
543 | if (err != 0) { | |
34dc7c2f | 544 | txh->txh_tx->tx_err = err; |
b128c09f | 545 | dbuf_rele(dbuf, FTAG); |
34dc7c2f BB |
546 | break; |
547 | } | |
548 | ||
b128c09f BB |
549 | bp = dbuf->db.db_data; |
550 | bp += blkoff; | |
551 | ||
552 | for (i = 0; i < tochk; i++) { | |
428870ff BB |
553 | if (dsl_dataset_block_freeable(ds, &bp[i], |
554 | bp[i].blk_birth)) { | |
b128c09f | 555 | dprintf_bp(&bp[i], "can free old%s", ""); |
428870ff | 556 | space += bp_get_dsize(spa, &bp[i]); |
b128c09f BB |
557 | } |
558 | unref += BP_GET_ASIZE(bp); | |
559 | } | |
560 | dbuf_rele(dbuf, FTAG); | |
561 | ||
34dc7c2f BB |
562 | blkid += tochk; |
563 | nblks -= tochk; | |
564 | } | |
565 | rw_exit(&dn->dn_struct_rwlock); | |
566 | ||
b128c09f BB |
567 | /* account for new level 1 indirect blocks that might show up */ |
568 | if (skipped > 0) { | |
569 | txh->txh_fudge += skipped << dn->dn_indblkshift; | |
570 | skipped = MIN(skipped, DMU_MAX_DELETEBLKCNT >> epbs); | |
571 | txh->txh_memory_tohold += skipped << dn->dn_indblkshift; | |
572 | } | |
34dc7c2f BB |
573 | txh->txh_space_tofree += space; |
574 | txh->txh_space_tounref += unref; | |
575 | } | |
576 | ||
577 | void | |
578 | dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, uint64_t len) | |
579 | { | |
580 | dmu_tx_hold_t *txh; | |
581 | dnode_t *dn; | |
582 | uint64_t start, end, i; | |
583 | int err, shift; | |
584 | zio_t *zio; | |
585 | ||
586 | ASSERT(tx->tx_txg == 0); | |
587 | ||
588 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, | |
589 | object, THT_FREE, off, len); | |
590 | if (txh == NULL) | |
591 | return; | |
592 | dn = txh->txh_dnode; | |
593 | ||
594 | /* first block */ | |
595 | if (off != 0) | |
596 | dmu_tx_count_write(txh, off, 1); | |
597 | /* last block */ | |
598 | if (len != DMU_OBJECT_END) | |
599 | dmu_tx_count_write(txh, off+len, 1); | |
600 | ||
428870ff BB |
601 | dmu_tx_count_dnode(txh); |
602 | ||
34dc7c2f BB |
603 | if (off >= (dn->dn_maxblkid+1) * dn->dn_datablksz) |
604 | return; | |
605 | if (len == DMU_OBJECT_END) | |
606 | len = (dn->dn_maxblkid+1) * dn->dn_datablksz - off; | |
607 | ||
608 | /* | |
609 | * For i/o error checking, read the first and last level-0 | |
610 | * blocks, and all the level-1 blocks. The above count_write's | |
b128c09f | 611 | * have already taken care of the level-0 blocks. |
34dc7c2f BB |
612 | */ |
613 | if (dn->dn_nlevels > 1) { | |
614 | shift = dn->dn_datablkshift + dn->dn_indblkshift - | |
615 | SPA_BLKPTRSHIFT; | |
616 | start = off >> shift; | |
617 | end = dn->dn_datablkshift ? ((off+len) >> shift) : 0; | |
618 | ||
619 | zio = zio_root(tx->tx_pool->dp_spa, | |
620 | NULL, NULL, ZIO_FLAG_CANFAIL); | |
621 | for (i = start; i <= end; i++) { | |
622 | uint64_t ibyte = i << shift; | |
b128c09f | 623 | err = dnode_next_offset(dn, 0, &ibyte, 2, 1, 0); |
34dc7c2f BB |
624 | i = ibyte >> shift; |
625 | if (err == ESRCH) | |
626 | break; | |
627 | if (err) { | |
628 | tx->tx_err = err; | |
629 | return; | |
630 | } | |
631 | ||
632 | err = dmu_tx_check_ioerr(zio, dn, 1, i); | |
633 | if (err) { | |
634 | tx->tx_err = err; | |
635 | return; | |
636 | } | |
637 | } | |
638 | err = zio_wait(zio); | |
639 | if (err) { | |
640 | tx->tx_err = err; | |
641 | return; | |
642 | } | |
643 | } | |
644 | ||
34dc7c2f BB |
645 | dmu_tx_count_free(txh, off, len); |
646 | } | |
647 | ||
648 | void | |
9babb374 | 649 | dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name) |
34dc7c2f BB |
650 | { |
651 | dmu_tx_hold_t *txh; | |
652 | dnode_t *dn; | |
653 | uint64_t nblocks; | |
654 | int epbs, err; | |
655 | ||
656 | ASSERT(tx->tx_txg == 0); | |
657 | ||
658 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, | |
659 | object, THT_ZAP, add, (uintptr_t)name); | |
660 | if (txh == NULL) | |
661 | return; | |
662 | dn = txh->txh_dnode; | |
663 | ||
664 | dmu_tx_count_dnode(txh); | |
665 | ||
666 | if (dn == NULL) { | |
667 | /* | |
668 | * We will be able to fit a new object's entries into one leaf | |
669 | * block. So there will be at most 2 blocks total, | |
670 | * including the header block. | |
671 | */ | |
672 | dmu_tx_count_write(txh, 0, 2 << fzap_default_block_shift); | |
673 | return; | |
674 | } | |
675 | ||
676 | ASSERT3P(dmu_ot[dn->dn_type].ot_byteswap, ==, zap_byteswap); | |
677 | ||
678 | if (dn->dn_maxblkid == 0 && !add) { | |
679 | /* | |
680 | * If there is only one block (i.e. this is a micro-zap) | |
681 | * and we are not adding anything, the accounting is simple. | |
682 | */ | |
683 | err = dmu_tx_check_ioerr(NULL, dn, 0, 0); | |
684 | if (err) { | |
685 | tx->tx_err = err; | |
686 | return; | |
687 | } | |
688 | ||
689 | /* | |
690 | * Use max block size here, since we don't know how much | |
691 | * the size will change between now and the dbuf dirty call. | |
692 | */ | |
693 | if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset, | |
428870ff | 694 | &dn->dn_phys->dn_blkptr[0], |
34dc7c2f BB |
695 | dn->dn_phys->dn_blkptr[0].blk_birth)) { |
696 | txh->txh_space_tooverwrite += SPA_MAXBLOCKSIZE; | |
697 | } else { | |
698 | txh->txh_space_towrite += SPA_MAXBLOCKSIZE; | |
34dc7c2f | 699 | } |
9babb374 BB |
700 | if (dn->dn_phys->dn_blkptr[0].blk_birth) |
701 | txh->txh_space_tounref += SPA_MAXBLOCKSIZE; | |
34dc7c2f BB |
702 | return; |
703 | } | |
704 | ||
705 | if (dn->dn_maxblkid > 0 && name) { | |
706 | /* | |
707 | * access the name in this fat-zap so that we'll check | |
708 | * for i/o errors to the leaf blocks, etc. | |
709 | */ | |
428870ff | 710 | err = zap_lookup(dn->dn_objset, dn->dn_object, name, |
34dc7c2f BB |
711 | 8, 0, NULL); |
712 | if (err == EIO) { | |
713 | tx->tx_err = err; | |
714 | return; | |
715 | } | |
716 | } | |
717 | ||
428870ff | 718 | err = zap_count_write(dn->dn_objset, dn->dn_object, name, add, |
45d1cae3 | 719 | &txh->txh_space_towrite, &txh->txh_space_tooverwrite); |
34dc7c2f BB |
720 | |
721 | /* | |
722 | * If the modified blocks are scattered to the four winds, | |
723 | * we'll have to modify an indirect twig for each. | |
724 | */ | |
725 | epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
726 | for (nblocks = dn->dn_maxblkid >> epbs; nblocks != 0; nblocks >>= epbs) | |
9babb374 BB |
727 | if (dn->dn_objset->os_dsl_dataset->ds_phys->ds_prev_snap_obj) |
728 | txh->txh_space_towrite += 3 << dn->dn_indblkshift; | |
729 | else | |
730 | txh->txh_space_tooverwrite += 3 << dn->dn_indblkshift; | |
34dc7c2f BB |
731 | } |
732 | ||
733 | void | |
734 | dmu_tx_hold_bonus(dmu_tx_t *tx, uint64_t object) | |
735 | { | |
736 | dmu_tx_hold_t *txh; | |
737 | ||
738 | ASSERT(tx->tx_txg == 0); | |
739 | ||
740 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, | |
741 | object, THT_BONUS, 0, 0); | |
742 | if (txh) | |
743 | dmu_tx_count_dnode(txh); | |
744 | } | |
745 | ||
746 | void | |
747 | dmu_tx_hold_space(dmu_tx_t *tx, uint64_t space) | |
748 | { | |
749 | dmu_tx_hold_t *txh; | |
750 | ASSERT(tx->tx_txg == 0); | |
751 | ||
752 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, | |
753 | DMU_NEW_OBJECT, THT_SPACE, space, 0); | |
754 | ||
755 | txh->txh_space_towrite += space; | |
756 | } | |
757 | ||
758 | int | |
759 | dmu_tx_holds(dmu_tx_t *tx, uint64_t object) | |
760 | { | |
761 | dmu_tx_hold_t *txh; | |
762 | int holds = 0; | |
763 | ||
764 | /* | |
765 | * By asserting that the tx is assigned, we're counting the | |
766 | * number of dn_tx_holds, which is the same as the number of | |
767 | * dn_holds. Otherwise, we'd be counting dn_holds, but | |
768 | * dn_tx_holds could be 0. | |
769 | */ | |
770 | ASSERT(tx->tx_txg != 0); | |
771 | ||
772 | /* if (tx->tx_anyobj == TRUE) */ | |
773 | /* return (0); */ | |
774 | ||
775 | for (txh = list_head(&tx->tx_holds); txh; | |
776 | txh = list_next(&tx->tx_holds, txh)) { | |
777 | if (txh->txh_dnode && txh->txh_dnode->dn_object == object) | |
778 | holds++; | |
779 | } | |
780 | ||
781 | return (holds); | |
782 | } | |
783 | ||
784 | #ifdef ZFS_DEBUG | |
785 | void | |
786 | dmu_tx_dirty_buf(dmu_tx_t *tx, dmu_buf_impl_t *db) | |
787 | { | |
788 | dmu_tx_hold_t *txh; | |
789 | int match_object = FALSE, match_offset = FALSE; | |
790 | dnode_t *dn = db->db_dnode; | |
791 | ||
792 | ASSERT(tx->tx_txg != 0); | |
428870ff | 793 | ASSERT(tx->tx_objset == NULL || dn->dn_objset == tx->tx_objset); |
34dc7c2f BB |
794 | ASSERT3U(dn->dn_object, ==, db->db.db_object); |
795 | ||
796 | if (tx->tx_anyobj) | |
797 | return; | |
798 | ||
799 | /* XXX No checking on the meta dnode for now */ | |
800 | if (db->db.db_object == DMU_META_DNODE_OBJECT) | |
801 | return; | |
802 | ||
803 | for (txh = list_head(&tx->tx_holds); txh; | |
804 | txh = list_next(&tx->tx_holds, txh)) { | |
805 | ASSERT(dn == NULL || dn->dn_assigned_txg == tx->tx_txg); | |
806 | if (txh->txh_dnode == dn && txh->txh_type != THT_NEWOBJECT) | |
807 | match_object = TRUE; | |
808 | if (txh->txh_dnode == NULL || txh->txh_dnode == dn) { | |
809 | int datablkshift = dn->dn_datablkshift ? | |
810 | dn->dn_datablkshift : SPA_MAXBLOCKSHIFT; | |
811 | int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
812 | int shift = datablkshift + epbs * db->db_level; | |
813 | uint64_t beginblk = shift >= 64 ? 0 : | |
814 | (txh->txh_arg1 >> shift); | |
815 | uint64_t endblk = shift >= 64 ? 0 : | |
816 | ((txh->txh_arg1 + txh->txh_arg2 - 1) >> shift); | |
817 | uint64_t blkid = db->db_blkid; | |
818 | ||
819 | /* XXX txh_arg2 better not be zero... */ | |
820 | ||
821 | dprintf("found txh type %x beginblk=%llx endblk=%llx\n", | |
822 | txh->txh_type, beginblk, endblk); | |
823 | ||
824 | switch (txh->txh_type) { | |
825 | case THT_WRITE: | |
826 | if (blkid >= beginblk && blkid <= endblk) | |
827 | match_offset = TRUE; | |
828 | /* | |
829 | * We will let this hold work for the bonus | |
428870ff BB |
830 | * or spill buffer so that we don't need to |
831 | * hold it when creating a new object. | |
34dc7c2f | 832 | */ |
428870ff BB |
833 | if (blkid == DMU_BONUS_BLKID || |
834 | blkid == DMU_SPILL_BLKID) | |
34dc7c2f BB |
835 | match_offset = TRUE; |
836 | /* | |
837 | * They might have to increase nlevels, | |
838 | * thus dirtying the new TLIBs. Or the | |
839 | * might have to change the block size, | |
840 | * thus dirying the new lvl=0 blk=0. | |
841 | */ | |
842 | if (blkid == 0) | |
843 | match_offset = TRUE; | |
844 | break; | |
845 | case THT_FREE: | |
b128c09f BB |
846 | /* |
847 | * We will dirty all the level 1 blocks in | |
848 | * the free range and perhaps the first and | |
849 | * last level 0 block. | |
850 | */ | |
851 | if (blkid >= beginblk && (blkid <= endblk || | |
852 | txh->txh_arg2 == DMU_OBJECT_END)) | |
34dc7c2f BB |
853 | match_offset = TRUE; |
854 | break; | |
428870ff BB |
855 | case THT_SPILL: |
856 | if (blkid == DMU_SPILL_BLKID) | |
857 | match_offset = TRUE; | |
858 | break; | |
34dc7c2f | 859 | case THT_BONUS: |
428870ff | 860 | if (blkid == DMU_BONUS_BLKID) |
34dc7c2f BB |
861 | match_offset = TRUE; |
862 | break; | |
863 | case THT_ZAP: | |
864 | match_offset = TRUE; | |
865 | break; | |
866 | case THT_NEWOBJECT: | |
867 | match_object = TRUE; | |
868 | break; | |
869 | default: | |
870 | ASSERT(!"bad txh_type"); | |
871 | } | |
872 | } | |
873 | if (match_object && match_offset) | |
874 | return; | |
875 | } | |
876 | panic("dirtying dbuf obj=%llx lvl=%u blkid=%llx but not tx_held\n", | |
877 | (u_longlong_t)db->db.db_object, db->db_level, | |
878 | (u_longlong_t)db->db_blkid); | |
879 | } | |
880 | #endif | |
881 | ||
882 | static int | |
883 | dmu_tx_try_assign(dmu_tx_t *tx, uint64_t txg_how) | |
884 | { | |
885 | dmu_tx_hold_t *txh; | |
886 | spa_t *spa = tx->tx_pool->dp_spa; | |
b128c09f BB |
887 | uint64_t memory, asize, fsize, usize; |
888 | uint64_t towrite, tofree, tooverwrite, tounref, tohold, fudge; | |
34dc7c2f BB |
889 | |
890 | ASSERT3U(tx->tx_txg, ==, 0); | |
891 | ||
892 | if (tx->tx_err) | |
893 | return (tx->tx_err); | |
894 | ||
b128c09f | 895 | if (spa_suspended(spa)) { |
34dc7c2f BB |
896 | /* |
897 | * If the user has indicated a blocking failure mode | |
898 | * then return ERESTART which will block in dmu_tx_wait(). | |
899 | * Otherwise, return EIO so that an error can get | |
900 | * propagated back to the VOP calls. | |
901 | * | |
902 | * Note that we always honor the txg_how flag regardless | |
903 | * of the failuremode setting. | |
904 | */ | |
905 | if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE && | |
906 | txg_how != TXG_WAIT) | |
907 | return (EIO); | |
908 | ||
909 | return (ERESTART); | |
910 | } | |
911 | ||
912 | tx->tx_txg = txg_hold_open(tx->tx_pool, &tx->tx_txgh); | |
913 | tx->tx_needassign_txh = NULL; | |
914 | ||
915 | /* | |
916 | * NB: No error returns are allowed after txg_hold_open, but | |
917 | * before processing the dnode holds, due to the | |
918 | * dmu_tx_unassign() logic. | |
919 | */ | |
920 | ||
b128c09f | 921 | towrite = tofree = tooverwrite = tounref = tohold = fudge = 0; |
34dc7c2f BB |
922 | for (txh = list_head(&tx->tx_holds); txh; |
923 | txh = list_next(&tx->tx_holds, txh)) { | |
924 | dnode_t *dn = txh->txh_dnode; | |
925 | if (dn != NULL) { | |
926 | mutex_enter(&dn->dn_mtx); | |
927 | if (dn->dn_assigned_txg == tx->tx_txg - 1) { | |
928 | mutex_exit(&dn->dn_mtx); | |
929 | tx->tx_needassign_txh = txh; | |
930 | return (ERESTART); | |
931 | } | |
932 | if (dn->dn_assigned_txg == 0) | |
933 | dn->dn_assigned_txg = tx->tx_txg; | |
934 | ASSERT3U(dn->dn_assigned_txg, ==, tx->tx_txg); | |
935 | (void) refcount_add(&dn->dn_tx_holds, tx); | |
936 | mutex_exit(&dn->dn_mtx); | |
937 | } | |
938 | towrite += txh->txh_space_towrite; | |
939 | tofree += txh->txh_space_tofree; | |
940 | tooverwrite += txh->txh_space_tooverwrite; | |
941 | tounref += txh->txh_space_tounref; | |
b128c09f BB |
942 | tohold += txh->txh_memory_tohold; |
943 | fudge += txh->txh_fudge; | |
34dc7c2f BB |
944 | } |
945 | ||
946 | /* | |
947 | * NB: This check must be after we've held the dnodes, so that | |
948 | * the dmu_tx_unassign() logic will work properly | |
949 | */ | |
950 | if (txg_how >= TXG_INITIAL && txg_how != tx->tx_txg) | |
951 | return (ERESTART); | |
952 | ||
953 | /* | |
954 | * If a snapshot has been taken since we made our estimates, | |
955 | * assume that we won't be able to free or overwrite anything. | |
956 | */ | |
957 | if (tx->tx_objset && | |
428870ff | 958 | dsl_dataset_prev_snap_txg(tx->tx_objset->os_dsl_dataset) > |
34dc7c2f BB |
959 | tx->tx_lastsnap_txg) { |
960 | towrite += tooverwrite; | |
961 | tooverwrite = tofree = 0; | |
962 | } | |
963 | ||
b128c09f BB |
964 | /* needed allocation: worst-case estimate of write space */ |
965 | asize = spa_get_asize(tx->tx_pool->dp_spa, towrite + tooverwrite); | |
966 | /* freed space estimate: worst-case overwrite + free estimate */ | |
34dc7c2f | 967 | fsize = spa_get_asize(tx->tx_pool->dp_spa, tooverwrite) + tofree; |
b128c09f | 968 | /* convert unrefd space to worst-case estimate */ |
34dc7c2f | 969 | usize = spa_get_asize(tx->tx_pool->dp_spa, tounref); |
b128c09f BB |
970 | /* calculate memory footprint estimate */ |
971 | memory = towrite + tooverwrite + tohold; | |
34dc7c2f BB |
972 | |
973 | #ifdef ZFS_DEBUG | |
b128c09f BB |
974 | /* |
975 | * Add in 'tohold' to account for our dirty holds on this memory | |
976 | * XXX - the "fudge" factor is to account for skipped blocks that | |
977 | * we missed because dnode_next_offset() misses in-core-only blocks. | |
978 | */ | |
979 | tx->tx_space_towrite = asize + | |
980 | spa_get_asize(tx->tx_pool->dp_spa, tohold + fudge); | |
34dc7c2f BB |
981 | tx->tx_space_tofree = tofree; |
982 | tx->tx_space_tooverwrite = tooverwrite; | |
983 | tx->tx_space_tounref = tounref; | |
984 | #endif | |
985 | ||
986 | if (tx->tx_dir && asize != 0) { | |
b128c09f BB |
987 | int err = dsl_dir_tempreserve_space(tx->tx_dir, memory, |
988 | asize, fsize, usize, &tx->tx_tempreserve_cookie, tx); | |
34dc7c2f BB |
989 | if (err) |
990 | return (err); | |
991 | } | |
992 | ||
993 | return (0); | |
994 | } | |
995 | ||
996 | static void | |
997 | dmu_tx_unassign(dmu_tx_t *tx) | |
998 | { | |
999 | dmu_tx_hold_t *txh; | |
1000 | ||
1001 | if (tx->tx_txg == 0) | |
1002 | return; | |
1003 | ||
1004 | txg_rele_to_quiesce(&tx->tx_txgh); | |
1005 | ||
1006 | for (txh = list_head(&tx->tx_holds); txh != tx->tx_needassign_txh; | |
1007 | txh = list_next(&tx->tx_holds, txh)) { | |
1008 | dnode_t *dn = txh->txh_dnode; | |
1009 | ||
1010 | if (dn == NULL) | |
1011 | continue; | |
1012 | mutex_enter(&dn->dn_mtx); | |
1013 | ASSERT3U(dn->dn_assigned_txg, ==, tx->tx_txg); | |
1014 | ||
1015 | if (refcount_remove(&dn->dn_tx_holds, tx) == 0) { | |
1016 | dn->dn_assigned_txg = 0; | |
1017 | cv_broadcast(&dn->dn_notxholds); | |
1018 | } | |
1019 | mutex_exit(&dn->dn_mtx); | |
1020 | } | |
1021 | ||
1022 | txg_rele_to_sync(&tx->tx_txgh); | |
1023 | ||
1024 | tx->tx_lasttried_txg = tx->tx_txg; | |
1025 | tx->tx_txg = 0; | |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * Assign tx to a transaction group. txg_how can be one of: | |
1030 | * | |
1031 | * (1) TXG_WAIT. If the current open txg is full, waits until there's | |
1032 | * a new one. This should be used when you're not holding locks. | |
1033 | * If will only fail if we're truly out of space (or over quota). | |
1034 | * | |
1035 | * (2) TXG_NOWAIT. If we can't assign into the current open txg without | |
1036 | * blocking, returns immediately with ERESTART. This should be used | |
1037 | * whenever you're holding locks. On an ERESTART error, the caller | |
1038 | * should drop locks, do a dmu_tx_wait(tx), and try again. | |
1039 | * | |
1040 | * (3) A specific txg. Use this if you need to ensure that multiple | |
1041 | * transactions all sync in the same txg. Like TXG_NOWAIT, it | |
1042 | * returns ERESTART if it can't assign you into the requested txg. | |
1043 | */ | |
1044 | int | |
1045 | dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how) | |
1046 | { | |
1047 | int err; | |
1048 | ||
1049 | ASSERT(tx->tx_txg == 0); | |
1050 | ASSERT(txg_how != 0); | |
1051 | ASSERT(!dsl_pool_sync_context(tx->tx_pool)); | |
1052 | ||
1053 | while ((err = dmu_tx_try_assign(tx, txg_how)) != 0) { | |
1054 | dmu_tx_unassign(tx); | |
1055 | ||
1056 | if (err != ERESTART || txg_how != TXG_WAIT) | |
1057 | return (err); | |
1058 | ||
1059 | dmu_tx_wait(tx); | |
1060 | } | |
1061 | ||
1062 | txg_rele_to_quiesce(&tx->tx_txgh); | |
1063 | ||
1064 | return (0); | |
1065 | } | |
1066 | ||
1067 | void | |
1068 | dmu_tx_wait(dmu_tx_t *tx) | |
1069 | { | |
1070 | spa_t *spa = tx->tx_pool->dp_spa; | |
1071 | ||
1072 | ASSERT(tx->tx_txg == 0); | |
1073 | ||
1074 | /* | |
1075 | * It's possible that the pool has become active after this thread | |
1076 | * has tried to obtain a tx. If that's the case then his | |
1077 | * tx_lasttried_txg would not have been assigned. | |
1078 | */ | |
b128c09f | 1079 | if (spa_suspended(spa) || tx->tx_lasttried_txg == 0) { |
34dc7c2f BB |
1080 | txg_wait_synced(tx->tx_pool, spa_last_synced_txg(spa) + 1); |
1081 | } else if (tx->tx_needassign_txh) { | |
1082 | dnode_t *dn = tx->tx_needassign_txh->txh_dnode; | |
1083 | ||
1084 | mutex_enter(&dn->dn_mtx); | |
1085 | while (dn->dn_assigned_txg == tx->tx_lasttried_txg - 1) | |
1086 | cv_wait(&dn->dn_notxholds, &dn->dn_mtx); | |
1087 | mutex_exit(&dn->dn_mtx); | |
1088 | tx->tx_needassign_txh = NULL; | |
1089 | } else { | |
1090 | txg_wait_open(tx->tx_pool, tx->tx_lasttried_txg + 1); | |
1091 | } | |
1092 | } | |
1093 | ||
1094 | void | |
1095 | dmu_tx_willuse_space(dmu_tx_t *tx, int64_t delta) | |
1096 | { | |
1097 | #ifdef ZFS_DEBUG | |
1098 | if (tx->tx_dir == NULL || delta == 0) | |
1099 | return; | |
1100 | ||
1101 | if (delta > 0) { | |
1102 | ASSERT3U(refcount_count(&tx->tx_space_written) + delta, <=, | |
1103 | tx->tx_space_towrite); | |
1104 | (void) refcount_add_many(&tx->tx_space_written, delta, NULL); | |
1105 | } else { | |
1106 | (void) refcount_add_many(&tx->tx_space_freed, -delta, NULL); | |
1107 | } | |
1108 | #endif | |
1109 | } | |
1110 | ||
1111 | void | |
1112 | dmu_tx_commit(dmu_tx_t *tx) | |
1113 | { | |
1114 | dmu_tx_hold_t *txh; | |
1115 | ||
1116 | ASSERT(tx->tx_txg != 0); | |
1117 | ||
1118 | while (txh = list_head(&tx->tx_holds)) { | |
1119 | dnode_t *dn = txh->txh_dnode; | |
1120 | ||
1121 | list_remove(&tx->tx_holds, txh); | |
1122 | kmem_free(txh, sizeof (dmu_tx_hold_t)); | |
1123 | if (dn == NULL) | |
1124 | continue; | |
1125 | mutex_enter(&dn->dn_mtx); | |
1126 | ASSERT3U(dn->dn_assigned_txg, ==, tx->tx_txg); | |
1127 | ||
1128 | if (refcount_remove(&dn->dn_tx_holds, tx) == 0) { | |
1129 | dn->dn_assigned_txg = 0; | |
1130 | cv_broadcast(&dn->dn_notxholds); | |
1131 | } | |
1132 | mutex_exit(&dn->dn_mtx); | |
1133 | dnode_rele(dn, tx); | |
1134 | } | |
1135 | ||
1136 | if (tx->tx_tempreserve_cookie) | |
1137 | dsl_dir_tempreserve_clear(tx->tx_tempreserve_cookie, tx); | |
1138 | ||
428870ff BB |
1139 | if (!list_is_empty(&tx->tx_callbacks)) |
1140 | txg_register_callbacks(&tx->tx_txgh, &tx->tx_callbacks); | |
1141 | ||
34dc7c2f BB |
1142 | if (tx->tx_anyobj == FALSE) |
1143 | txg_rele_to_sync(&tx->tx_txgh); | |
428870ff BB |
1144 | |
1145 | list_destroy(&tx->tx_callbacks); | |
34dc7c2f BB |
1146 | list_destroy(&tx->tx_holds); |
1147 | #ifdef ZFS_DEBUG | |
1148 | dprintf("towrite=%llu written=%llu tofree=%llu freed=%llu\n", | |
1149 | tx->tx_space_towrite, refcount_count(&tx->tx_space_written), | |
1150 | tx->tx_space_tofree, refcount_count(&tx->tx_space_freed)); | |
1151 | refcount_destroy_many(&tx->tx_space_written, | |
1152 | refcount_count(&tx->tx_space_written)); | |
1153 | refcount_destroy_many(&tx->tx_space_freed, | |
1154 | refcount_count(&tx->tx_space_freed)); | |
1155 | #endif | |
1156 | kmem_free(tx, sizeof (dmu_tx_t)); | |
1157 | } | |
1158 | ||
1159 | void | |
1160 | dmu_tx_abort(dmu_tx_t *tx) | |
1161 | { | |
1162 | dmu_tx_hold_t *txh; | |
1163 | ||
1164 | ASSERT(tx->tx_txg == 0); | |
1165 | ||
1166 | while (txh = list_head(&tx->tx_holds)) { | |
1167 | dnode_t *dn = txh->txh_dnode; | |
1168 | ||
1169 | list_remove(&tx->tx_holds, txh); | |
1170 | kmem_free(txh, sizeof (dmu_tx_hold_t)); | |
1171 | if (dn != NULL) | |
1172 | dnode_rele(dn, tx); | |
1173 | } | |
428870ff BB |
1174 | |
1175 | /* | |
1176 | * Call any registered callbacks with an error code. | |
1177 | */ | |
1178 | if (!list_is_empty(&tx->tx_callbacks)) | |
1179 | dmu_tx_do_callbacks(&tx->tx_callbacks, ECANCELED); | |
1180 | ||
1181 | list_destroy(&tx->tx_callbacks); | |
34dc7c2f BB |
1182 | list_destroy(&tx->tx_holds); |
1183 | #ifdef ZFS_DEBUG | |
1184 | refcount_destroy_many(&tx->tx_space_written, | |
1185 | refcount_count(&tx->tx_space_written)); | |
1186 | refcount_destroy_many(&tx->tx_space_freed, | |
1187 | refcount_count(&tx->tx_space_freed)); | |
1188 | #endif | |
1189 | kmem_free(tx, sizeof (dmu_tx_t)); | |
1190 | } | |
1191 | ||
1192 | uint64_t | |
1193 | dmu_tx_get_txg(dmu_tx_t *tx) | |
1194 | { | |
1195 | ASSERT(tx->tx_txg != 0); | |
1196 | return (tx->tx_txg); | |
1197 | } | |
428870ff BB |
1198 | |
1199 | void | |
1200 | dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *func, void *data) | |
1201 | { | |
1202 | dmu_tx_callback_t *dcb; | |
1203 | ||
1204 | dcb = kmem_alloc(sizeof (dmu_tx_callback_t), KM_SLEEP); | |
1205 | ||
1206 | dcb->dcb_func = func; | |
1207 | dcb->dcb_data = data; | |
1208 | ||
1209 | list_insert_tail(&tx->tx_callbacks, dcb); | |
1210 | } | |
1211 | ||
1212 | /* | |
1213 | * Call all the commit callbacks on a list, with a given error code. | |
1214 | */ | |
1215 | void | |
1216 | dmu_tx_do_callbacks(list_t *cb_list, int error) | |
1217 | { | |
1218 | dmu_tx_callback_t *dcb; | |
1219 | ||
1220 | while (dcb = list_head(cb_list)) { | |
1221 | list_remove(cb_list, dcb); | |
1222 | dcb->dcb_func(dcb->dcb_data, error); | |
1223 | kmem_free(dcb, sizeof (dmu_tx_callback_t)); | |
1224 | } | |
1225 | } | |
1226 | ||
1227 | /* | |
1228 | * Interface to hold a bunch of attributes. | |
1229 | * used for creating new files. | |
1230 | * attrsize is the total size of all attributes | |
1231 | * to be added during object creation | |
1232 | * | |
1233 | * For updating/adding a single attribute dmu_tx_hold_sa() should be used. | |
1234 | */ | |
1235 | ||
1236 | /* | |
1237 | * hold necessary attribute name for attribute registration. | |
1238 | * should be a very rare case where this is needed. If it does | |
1239 | * happen it would only happen on the first write to the file system. | |
1240 | */ | |
1241 | static void | |
1242 | dmu_tx_sa_registration_hold(sa_os_t *sa, dmu_tx_t *tx) | |
1243 | { | |
1244 | int i; | |
1245 | ||
1246 | if (!sa->sa_need_attr_registration) | |
1247 | return; | |
1248 | ||
1249 | for (i = 0; i != sa->sa_num_attrs; i++) { | |
1250 | if (!sa->sa_attr_table[i].sa_registered) { | |
1251 | if (sa->sa_reg_attr_obj) | |
1252 | dmu_tx_hold_zap(tx, sa->sa_reg_attr_obj, | |
1253 | B_TRUE, sa->sa_attr_table[i].sa_name); | |
1254 | else | |
1255 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, | |
1256 | B_TRUE, sa->sa_attr_table[i].sa_name); | |
1257 | } | |
1258 | } | |
1259 | } | |
1260 | ||
1261 | ||
1262 | void | |
1263 | dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object) | |
1264 | { | |
1265 | dnode_t *dn; | |
1266 | dmu_tx_hold_t *txh; | |
1267 | blkptr_t *bp; | |
1268 | ||
1269 | txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, object, | |
1270 | THT_SPILL, 0, 0); | |
1271 | ||
1272 | dn = txh->txh_dnode; | |
1273 | ||
1274 | if (dn == NULL) | |
1275 | return; | |
1276 | ||
1277 | /* If blkptr doesn't exist then add space to towrite */ | |
1278 | bp = &dn->dn_phys->dn_spill; | |
1279 | if (BP_IS_HOLE(bp)) { | |
1280 | txh->txh_space_towrite += SPA_MAXBLOCKSIZE; | |
1281 | txh->txh_space_tounref = 0; | |
1282 | } else { | |
1283 | if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset, | |
1284 | bp, bp->blk_birth)) | |
1285 | txh->txh_space_tooverwrite += SPA_MAXBLOCKSIZE; | |
1286 | else | |
1287 | txh->txh_space_towrite += SPA_MAXBLOCKSIZE; | |
1288 | if (bp->blk_birth) | |
1289 | txh->txh_space_tounref += SPA_MAXBLOCKSIZE; | |
1290 | } | |
1291 | } | |
1292 | ||
1293 | void | |
1294 | dmu_tx_hold_sa_create(dmu_tx_t *tx, int attrsize) | |
1295 | { | |
1296 | sa_os_t *sa = tx->tx_objset->os_sa; | |
1297 | ||
1298 | dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); | |
1299 | ||
1300 | if (tx->tx_objset->os_sa->sa_master_obj == 0) | |
1301 | return; | |
1302 | ||
1303 | if (tx->tx_objset->os_sa->sa_layout_attr_obj) | |
1304 | dmu_tx_hold_zap(tx, sa->sa_layout_attr_obj, B_TRUE, NULL); | |
1305 | else { | |
1306 | dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_LAYOUTS); | |
1307 | dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_REGISTRY); | |
1308 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1309 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1310 | } | |
1311 | ||
1312 | dmu_tx_sa_registration_hold(sa, tx); | |
1313 | ||
1314 | if (attrsize <= DN_MAX_BONUSLEN && !sa->sa_force_spill) | |
1315 | return; | |
1316 | ||
1317 | (void) dmu_tx_hold_object_impl(tx, tx->tx_objset, DMU_NEW_OBJECT, | |
1318 | THT_SPILL, 0, 0); | |
1319 | } | |
1320 | ||
1321 | /* | |
1322 | * Hold SA attribute | |
1323 | * | |
1324 | * dmu_tx_hold_sa(dmu_tx_t *tx, sa_handle_t *, attribute, add, size) | |
1325 | * | |
1326 | * variable_size is the total size of all variable sized attributes | |
1327 | * passed to this function. It is not the total size of all | |
1328 | * variable size attributes that *may* exist on this object. | |
1329 | */ | |
1330 | void | |
1331 | dmu_tx_hold_sa(dmu_tx_t *tx, sa_handle_t *hdl, boolean_t may_grow) | |
1332 | { | |
1333 | uint64_t object; | |
1334 | sa_os_t *sa = tx->tx_objset->os_sa; | |
1335 | ||
1336 | ASSERT(hdl != NULL); | |
1337 | ||
1338 | object = sa_handle_object(hdl); | |
1339 | ||
1340 | dmu_tx_hold_bonus(tx, object); | |
1341 | ||
1342 | if (tx->tx_objset->os_sa->sa_master_obj == 0) | |
1343 | return; | |
1344 | ||
1345 | if (tx->tx_objset->os_sa->sa_reg_attr_obj == 0 || | |
1346 | tx->tx_objset->os_sa->sa_layout_attr_obj == 0) { | |
1347 | dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_LAYOUTS); | |
1348 | dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_REGISTRY); | |
1349 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1350 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL); | |
1351 | } | |
1352 | ||
1353 | dmu_tx_sa_registration_hold(sa, tx); | |
1354 | ||
1355 | if (may_grow && tx->tx_objset->os_sa->sa_layout_attr_obj) | |
1356 | dmu_tx_hold_zap(tx, sa->sa_layout_attr_obj, B_TRUE, NULL); | |
1357 | ||
1358 | if (sa->sa_force_spill || may_grow || hdl->sa_spill || | |
1359 | ((dmu_buf_impl_t *)hdl->sa_bonus)->db_dnode->dn_have_spill) { | |
1360 | ASSERT(tx->tx_txg == 0); | |
1361 | dmu_tx_hold_spill(tx, object); | |
1362 | } | |
1363 | } |