]>
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/dmu_tx.h> | |
28 | #include <sys/dbuf.h> | |
29 | #include <sys/dnode.h> | |
30 | #include <sys/zfs_context.h> | |
31 | #include <sys/dmu_objset.h> | |
32 | #include <sys/dmu_traverse.h> | |
33 | #include <sys/dsl_dataset.h> | |
34 | #include <sys/dsl_dir.h> | |
35 | #include <sys/dsl_pool.h> | |
36 | #include <sys/dsl_synctask.h> | |
37 | #include <sys/dsl_prop.h> | |
38 | #include <sys/dmu_zfetch.h> | |
39 | #include <sys/zfs_ioctl.h> | |
40 | #include <sys/zap.h> | |
41 | #include <sys/zio_checksum.h> | |
428870ff | 42 | #include <sys/sa.h> |
34dc7c2f BB |
43 | #ifdef _KERNEL |
44 | #include <sys/vmsystm.h> | |
b128c09f | 45 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
46 | #endif |
47 | ||
48 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { | |
49 | { byteswap_uint8_array, TRUE, "unallocated" }, | |
50 | { zap_byteswap, TRUE, "object directory" }, | |
51 | { byteswap_uint64_array, TRUE, "object array" }, | |
52 | { byteswap_uint8_array, TRUE, "packed nvlist" }, | |
53 | { byteswap_uint64_array, TRUE, "packed nvlist size" }, | |
428870ff BB |
54 | { byteswap_uint64_array, TRUE, "bpobj" }, |
55 | { byteswap_uint64_array, TRUE, "bpobj header" }, | |
34dc7c2f BB |
56 | { byteswap_uint64_array, TRUE, "SPA space map header" }, |
57 | { byteswap_uint64_array, TRUE, "SPA space map" }, | |
58 | { byteswap_uint64_array, TRUE, "ZIL intent log" }, | |
59 | { dnode_buf_byteswap, TRUE, "DMU dnode" }, | |
60 | { dmu_objset_byteswap, TRUE, "DMU objset" }, | |
61 | { byteswap_uint64_array, TRUE, "DSL directory" }, | |
62 | { zap_byteswap, TRUE, "DSL directory child map"}, | |
63 | { zap_byteswap, TRUE, "DSL dataset snap map" }, | |
64 | { zap_byteswap, TRUE, "DSL props" }, | |
65 | { byteswap_uint64_array, TRUE, "DSL dataset" }, | |
66 | { zfs_znode_byteswap, TRUE, "ZFS znode" }, | |
67 | { zfs_oldacl_byteswap, TRUE, "ZFS V0 ACL" }, | |
68 | { byteswap_uint8_array, FALSE, "ZFS plain file" }, | |
69 | { zap_byteswap, TRUE, "ZFS directory" }, | |
70 | { zap_byteswap, TRUE, "ZFS master node" }, | |
71 | { zap_byteswap, TRUE, "ZFS delete queue" }, | |
72 | { byteswap_uint8_array, FALSE, "zvol object" }, | |
73 | { zap_byteswap, TRUE, "zvol prop" }, | |
74 | { byteswap_uint8_array, FALSE, "other uint8[]" }, | |
75 | { byteswap_uint64_array, FALSE, "other uint64[]" }, | |
76 | { zap_byteswap, TRUE, "other ZAP" }, | |
77 | { zap_byteswap, TRUE, "persistent error log" }, | |
78 | { byteswap_uint8_array, TRUE, "SPA history" }, | |
79 | { byteswap_uint64_array, TRUE, "SPA history offsets" }, | |
80 | { zap_byteswap, TRUE, "Pool properties" }, | |
81 | { zap_byteswap, TRUE, "DSL permissions" }, | |
82 | { zfs_acl_byteswap, TRUE, "ZFS ACL" }, | |
83 | { byteswap_uint8_array, TRUE, "ZFS SYSACL" }, | |
84 | { byteswap_uint8_array, TRUE, "FUID table" }, | |
85 | { byteswap_uint64_array, TRUE, "FUID table size" }, | |
b128c09f | 86 | { zap_byteswap, TRUE, "DSL dataset next clones"}, |
428870ff | 87 | { zap_byteswap, TRUE, "scan work queue" }, |
9babb374 BB |
88 | { zap_byteswap, TRUE, "ZFS user/group used" }, |
89 | { zap_byteswap, TRUE, "ZFS user/group quota" }, | |
45d1cae3 | 90 | { zap_byteswap, TRUE, "snapshot refcount tags"}, |
428870ff BB |
91 | { zap_byteswap, TRUE, "DDT ZAP algorithm" }, |
92 | { zap_byteswap, TRUE, "DDT statistics" }, | |
93 | { byteswap_uint8_array, TRUE, "System attributes" }, | |
94 | { zap_byteswap, TRUE, "SA master node" }, | |
95 | { zap_byteswap, TRUE, "SA attr registration" }, | |
96 | { zap_byteswap, TRUE, "SA attr layouts" }, | |
97 | { zap_byteswap, TRUE, "scan translations" }, | |
98 | { byteswap_uint8_array, FALSE, "deduplicated block" }, | |
99 | { zap_byteswap, TRUE, "DSL deadlist map" }, | |
100 | { byteswap_uint64_array, TRUE, "DSL deadlist map hdr" }, | |
101 | { zap_byteswap, TRUE, "DSL dir clones" }, | |
102 | { byteswap_uint64_array, TRUE, "bpobj subobj" }, | |
34dc7c2f BB |
103 | }; |
104 | ||
105 | int | |
106 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
428870ff | 107 | void *tag, dmu_buf_t **dbp, int flags) |
34dc7c2f BB |
108 | { |
109 | dnode_t *dn; | |
110 | uint64_t blkid; | |
111 | dmu_buf_impl_t *db; | |
112 | int err; | |
428870ff | 113 | int db_flags = DB_RF_CANFAIL; |
34dc7c2f | 114 | |
428870ff BB |
115 | if (flags & DMU_READ_NO_PREFETCH) |
116 | db_flags |= DB_RF_NOPREFETCH; | |
117 | ||
118 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
119 | if (err) |
120 | return (err); | |
121 | blkid = dbuf_whichblock(dn, offset); | |
122 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
123 | db = dbuf_hold(dn, blkid, tag); | |
124 | rw_exit(&dn->dn_struct_rwlock); | |
125 | if (db == NULL) { | |
126 | err = EIO; | |
127 | } else { | |
428870ff | 128 | err = dbuf_read(db, NULL, db_flags); |
34dc7c2f BB |
129 | if (err) { |
130 | dbuf_rele(db, tag); | |
131 | db = NULL; | |
132 | } | |
133 | } | |
134 | ||
135 | dnode_rele(dn, FTAG); | |
572e2857 | 136 | *dbp = &db->db; /* NULL db plus first field offset is NULL */ |
34dc7c2f BB |
137 | return (err); |
138 | } | |
139 | ||
140 | int | |
141 | dmu_bonus_max(void) | |
142 | { | |
143 | return (DN_MAX_BONUSLEN); | |
144 | } | |
145 | ||
146 | int | |
572e2857 | 147 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 148 | { |
572e2857 BB |
149 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
150 | dnode_t *dn; | |
151 | int error; | |
34dc7c2f | 152 | |
572e2857 BB |
153 | DB_DNODE_ENTER(db); |
154 | dn = DB_DNODE(db); | |
155 | ||
156 | if (dn->dn_bonus != db) { | |
157 | error = EINVAL; | |
158 | } else if (newsize < 0 || newsize > db_fake->db_size) { | |
159 | error = EINVAL; | |
160 | } else { | |
161 | dnode_setbonuslen(dn, newsize, tx); | |
162 | error = 0; | |
163 | } | |
164 | ||
165 | DB_DNODE_EXIT(db); | |
166 | return (error); | |
34dc7c2f BB |
167 | } |
168 | ||
428870ff | 169 | int |
572e2857 | 170 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 171 | { |
572e2857 BB |
172 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
173 | dnode_t *dn; | |
174 | int error; | |
428870ff | 175 | |
572e2857 BB |
176 | DB_DNODE_ENTER(db); |
177 | dn = DB_DNODE(db); | |
428870ff | 178 | |
572e2857 BB |
179 | if (type > DMU_OT_NUMTYPES) { |
180 | error = EINVAL; | |
181 | } else if (dn->dn_bonus != db) { | |
182 | error = EINVAL; | |
183 | } else { | |
184 | dnode_setbonus_type(dn, type, tx); | |
185 | error = 0; | |
186 | } | |
428870ff | 187 | |
572e2857 BB |
188 | DB_DNODE_EXIT(db); |
189 | return (error); | |
190 | } | |
191 | ||
192 | dmu_object_type_t | |
193 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
194 | { | |
195 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
196 | dnode_t *dn; | |
197 | dmu_object_type_t type; | |
198 | ||
199 | DB_DNODE_ENTER(db); | |
200 | dn = DB_DNODE(db); | |
201 | type = dn->dn_bonustype; | |
202 | DB_DNODE_EXIT(db); | |
203 | ||
204 | return (type); | |
428870ff BB |
205 | } |
206 | ||
207 | int | |
208 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
209 | { | |
210 | dnode_t *dn; | |
211 | int error; | |
212 | ||
213 | error = dnode_hold(os, object, FTAG, &dn); | |
214 | dbuf_rm_spill(dn, tx); | |
215 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
216 | dnode_rm_spill(dn, tx); | |
217 | rw_exit(&dn->dn_struct_rwlock); | |
218 | dnode_rele(dn, FTAG); | |
219 | return (error); | |
220 | } | |
221 | ||
34dc7c2f BB |
222 | /* |
223 | * returns ENOENT, EIO, or 0. | |
224 | */ | |
225 | int | |
226 | dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) | |
227 | { | |
228 | dnode_t *dn; | |
229 | dmu_buf_impl_t *db; | |
230 | int error; | |
231 | ||
428870ff | 232 | error = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
233 | if (error) |
234 | return (error); | |
235 | ||
236 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
237 | if (dn->dn_bonus == NULL) { | |
238 | rw_exit(&dn->dn_struct_rwlock); | |
239 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
240 | if (dn->dn_bonus == NULL) | |
241 | dbuf_create_bonus(dn); | |
242 | } | |
243 | db = dn->dn_bonus; | |
34dc7c2f BB |
244 | |
245 | /* as long as the bonus buf is held, the dnode will be held */ | |
572e2857 | 246 | if (refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 247 | VERIFY(dnode_add_ref(dn, db)); |
572e2857 BB |
248 | (void) atomic_inc_32_nv(&dn->dn_dbufs_count); |
249 | } | |
250 | ||
251 | /* | |
252 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
253 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
254 | * a dnode hold for every dbuf. | |
255 | */ | |
256 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
257 | |
258 | dnode_rele(dn, FTAG); | |
259 | ||
428870ff | 260 | VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); |
34dc7c2f BB |
261 | |
262 | *dbp = &db->db; | |
263 | return (0); | |
264 | } | |
265 | ||
428870ff BB |
266 | /* |
267 | * returns ENOENT, EIO, or 0. | |
268 | * | |
269 | * This interface will allocate a blank spill dbuf when a spill blk | |
270 | * doesn't already exist on the dnode. | |
271 | * | |
272 | * if you only want to find an already existing spill db, then | |
273 | * dmu_spill_hold_existing() should be used. | |
274 | */ | |
275 | int | |
276 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
277 | { | |
278 | dmu_buf_impl_t *db = NULL; | |
279 | int err; | |
280 | ||
281 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
282 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
283 | ||
284 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
285 | ||
286 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
287 | rw_exit(&dn->dn_struct_rwlock); | |
288 | ||
289 | ASSERT(db != NULL); | |
572e2857 BB |
290 | err = dbuf_read(db, NULL, flags); |
291 | if (err == 0) | |
292 | *dbp = &db->db; | |
293 | else | |
294 | dbuf_rele(db, tag); | |
428870ff BB |
295 | return (err); |
296 | } | |
297 | ||
298 | int | |
299 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
300 | { | |
572e2857 BB |
301 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
302 | dnode_t *dn; | |
428870ff BB |
303 | int err; |
304 | ||
572e2857 BB |
305 | DB_DNODE_ENTER(db); |
306 | dn = DB_DNODE(db); | |
307 | ||
308 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
309 | err = EINVAL; | |
310 | } else { | |
311 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
312 | ||
313 | if (!dn->dn_have_spill) { | |
314 | err = ENOENT; | |
315 | } else { | |
316 | err = dmu_spill_hold_by_dnode(dn, | |
317 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
318 | } | |
428870ff | 319 | |
428870ff | 320 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 321 | } |
572e2857 BB |
322 | |
323 | DB_DNODE_EXIT(db); | |
428870ff BB |
324 | return (err); |
325 | } | |
326 | ||
327 | int | |
328 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
329 | { | |
572e2857 BB |
330 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
331 | dnode_t *dn; | |
332 | int err; | |
333 | ||
334 | DB_DNODE_ENTER(db); | |
335 | dn = DB_DNODE(db); | |
336 | err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); | |
337 | DB_DNODE_EXIT(db); | |
338 | ||
339 | return (err); | |
428870ff BB |
340 | } |
341 | ||
34dc7c2f BB |
342 | /* |
343 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
344 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
345 | * and can induce severe lock contention when writing to several files | |
346 | * whose dnodes are in the same block. | |
347 | */ | |
348 | static int | |
9babb374 BB |
349 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
350 | int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) | |
34dc7c2f | 351 | { |
b128c09f | 352 | dsl_pool_t *dp = NULL; |
34dc7c2f BB |
353 | dmu_buf_t **dbp; |
354 | uint64_t blkid, nblks, i; | |
9babb374 | 355 | uint32_t dbuf_flags; |
34dc7c2f BB |
356 | int err; |
357 | zio_t *zio; | |
d4ed6673 | 358 | hrtime_t start = 0; |
34dc7c2f BB |
359 | |
360 | ASSERT(length <= DMU_MAX_ACCESS); | |
361 | ||
45d1cae3 | 362 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT; |
9babb374 BB |
363 | if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz) |
364 | dbuf_flags |= DB_RF_NOPREFETCH; | |
34dc7c2f BB |
365 | |
366 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
367 | if (dn->dn_datablkshift) { | |
368 | int blkshift = dn->dn_datablkshift; | |
369 | nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - | |
370 | P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; | |
371 | } else { | |
372 | if (offset + length > dn->dn_datablksz) { | |
373 | zfs_panic_recover("zfs: accessing past end of object " | |
374 | "%llx/%llx (size=%u access=%llu+%llu)", | |
375 | (longlong_t)dn->dn_objset-> | |
376 | os_dsl_dataset->ds_object, | |
377 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
378 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 379 | rw_exit(&dn->dn_struct_rwlock); |
34dc7c2f BB |
380 | return (EIO); |
381 | } | |
382 | nblks = 1; | |
383 | } | |
4db77a74 | 384 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP | KM_NODEBUG); |
34dc7c2f | 385 | |
b128c09f BB |
386 | if (dn->dn_objset->os_dsl_dataset) |
387 | dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool; | |
388 | if (dp && dsl_pool_sync_context(dp)) | |
389 | start = gethrtime(); | |
390 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); | |
34dc7c2f BB |
391 | blkid = dbuf_whichblock(dn, offset); |
392 | for (i = 0; i < nblks; i++) { | |
393 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); | |
394 | if (db == NULL) { | |
395 | rw_exit(&dn->dn_struct_rwlock); | |
396 | dmu_buf_rele_array(dbp, nblks, tag); | |
397 | zio_nowait(zio); | |
398 | return (EIO); | |
399 | } | |
400 | /* initiate async i/o */ | |
401 | if (read) { | |
9babb374 | 402 | (void) dbuf_read(db, zio, dbuf_flags); |
34dc7c2f BB |
403 | } |
404 | dbp[i] = &db->db; | |
405 | } | |
406 | rw_exit(&dn->dn_struct_rwlock); | |
407 | ||
408 | /* wait for async i/o */ | |
409 | err = zio_wait(zio); | |
b128c09f BB |
410 | /* track read overhead when we are in sync context */ |
411 | if (dp && dsl_pool_sync_context(dp)) | |
412 | dp->dp_read_overhead += gethrtime() - start; | |
34dc7c2f BB |
413 | if (err) { |
414 | dmu_buf_rele_array(dbp, nblks, tag); | |
415 | return (err); | |
416 | } | |
417 | ||
418 | /* wait for other io to complete */ | |
419 | if (read) { | |
420 | for (i = 0; i < nblks; i++) { | |
421 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
422 | mutex_enter(&db->db_mtx); | |
423 | while (db->db_state == DB_READ || | |
424 | db->db_state == DB_FILL) | |
425 | cv_wait(&db->db_changed, &db->db_mtx); | |
426 | if (db->db_state == DB_UNCACHED) | |
427 | err = EIO; | |
428 | mutex_exit(&db->db_mtx); | |
429 | if (err) { | |
430 | dmu_buf_rele_array(dbp, nblks, tag); | |
431 | return (err); | |
432 | } | |
433 | } | |
434 | } | |
435 | ||
436 | *numbufsp = nblks; | |
437 | *dbpp = dbp; | |
438 | return (0); | |
439 | } | |
440 | ||
441 | static int | |
442 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
443 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
444 | { | |
445 | dnode_t *dn; | |
446 | int err; | |
447 | ||
428870ff | 448 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
449 | if (err) |
450 | return (err); | |
451 | ||
452 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 453 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
454 | |
455 | dnode_rele(dn, FTAG); | |
456 | ||
457 | return (err); | |
458 | } | |
459 | ||
460 | int | |
572e2857 | 461 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
34dc7c2f BB |
462 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) |
463 | { | |
572e2857 BB |
464 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
465 | dnode_t *dn; | |
34dc7c2f BB |
466 | int err; |
467 | ||
572e2857 BB |
468 | DB_DNODE_ENTER(db); |
469 | dn = DB_DNODE(db); | |
34dc7c2f | 470 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 471 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 472 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
473 | |
474 | return (err); | |
475 | } | |
476 | ||
477 | void | |
478 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
479 | { | |
480 | int i; | |
481 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
482 | ||
483 | if (numbufs == 0) | |
484 | return; | |
485 | ||
486 | for (i = 0; i < numbufs; i++) { | |
487 | if (dbp[i]) | |
488 | dbuf_rele(dbp[i], tag); | |
489 | } | |
490 | ||
491 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
492 | } | |
493 | ||
494 | void | |
495 | dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) | |
496 | { | |
497 | dnode_t *dn; | |
498 | uint64_t blkid; | |
499 | int nblks, i, err; | |
500 | ||
501 | if (zfs_prefetch_disable) | |
502 | return; | |
503 | ||
504 | if (len == 0) { /* they're interested in the bonus buffer */ | |
572e2857 | 505 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
506 | |
507 | if (object == 0 || object >= DN_MAX_OBJECT) | |
508 | return; | |
509 | ||
510 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
511 | blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); | |
512 | dbuf_prefetch(dn, blkid); | |
513 | rw_exit(&dn->dn_struct_rwlock); | |
514 | return; | |
515 | } | |
516 | ||
517 | /* | |
518 | * XXX - Note, if the dnode for the requested object is not | |
519 | * already cached, we will do a *synchronous* read in the | |
520 | * dnode_hold() call. The same is true for any indirects. | |
521 | */ | |
428870ff | 522 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
523 | if (err != 0) |
524 | return; | |
525 | ||
526 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
527 | if (dn->dn_datablkshift) { | |
528 | int blkshift = dn->dn_datablkshift; | |
529 | nblks = (P2ROUNDUP(offset+len, 1<<blkshift) - | |
530 | P2ALIGN(offset, 1<<blkshift)) >> blkshift; | |
531 | } else { | |
532 | nblks = (offset < dn->dn_datablksz); | |
533 | } | |
534 | ||
535 | if (nblks != 0) { | |
536 | blkid = dbuf_whichblock(dn, offset); | |
537 | for (i = 0; i < nblks; i++) | |
538 | dbuf_prefetch(dn, blkid+i); | |
539 | } | |
540 | ||
541 | rw_exit(&dn->dn_struct_rwlock); | |
542 | ||
543 | dnode_rele(dn, FTAG); | |
544 | } | |
545 | ||
45d1cae3 BB |
546 | /* |
547 | * Get the next "chunk" of file data to free. We traverse the file from | |
548 | * the end so that the file gets shorter over time (if we crashes in the | |
549 | * middle, this will leave us in a better state). We find allocated file | |
550 | * data by simply searching the allocated level 1 indirects. | |
551 | */ | |
b128c09f | 552 | static int |
45d1cae3 | 553 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit) |
b128c09f | 554 | { |
45d1cae3 BB |
555 | uint64_t len = *start - limit; |
556 | uint64_t blkcnt = 0; | |
557 | uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1)); | |
558 | uint64_t iblkrange = | |
b128c09f BB |
559 | dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); |
560 | ||
45d1cae3 | 561 | ASSERT(limit <= *start); |
b128c09f | 562 | |
45d1cae3 BB |
563 | if (len <= iblkrange * maxblks) { |
564 | *start = limit; | |
b128c09f BB |
565 | return (0); |
566 | } | |
45d1cae3 | 567 | ASSERT(ISP2(iblkrange)); |
b128c09f | 568 | |
45d1cae3 | 569 | while (*start > limit && blkcnt < maxblks) { |
b128c09f BB |
570 | int err; |
571 | ||
45d1cae3 | 572 | /* find next allocated L1 indirect */ |
b128c09f | 573 | err = dnode_next_offset(dn, |
45d1cae3 | 574 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 575 | |
45d1cae3 BB |
576 | /* if there are no more, then we are done */ |
577 | if (err == ESRCH) { | |
578 | *start = limit; | |
b128c09f | 579 | return (0); |
45d1cae3 | 580 | } else if (err) { |
b128c09f | 581 | return (err); |
45d1cae3 BB |
582 | } |
583 | blkcnt += 1; | |
b128c09f | 584 | |
45d1cae3 BB |
585 | /* reset offset to end of "next" block back */ |
586 | *start = P2ALIGN(*start, iblkrange); | |
587 | if (*start <= limit) | |
588 | *start = limit; | |
589 | else | |
590 | *start -= 1; | |
b128c09f BB |
591 | } |
592 | return (0); | |
593 | } | |
594 | ||
595 | static int | |
596 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
597 | uint64_t length, boolean_t free_dnode) | |
598 | { | |
599 | dmu_tx_t *tx; | |
600 | uint64_t object_size, start, end, len; | |
601 | boolean_t trunc = (length == DMU_OBJECT_END); | |
602 | int align, err; | |
603 | ||
604 | align = 1 << dn->dn_datablkshift; | |
605 | ASSERT(align > 0); | |
606 | object_size = align == 1 ? dn->dn_datablksz : | |
607 | (dn->dn_maxblkid + 1) << dn->dn_datablkshift; | |
608 | ||
9babb374 BB |
609 | end = offset + length; |
610 | if (trunc || end > object_size) | |
b128c09f BB |
611 | end = object_size; |
612 | if (end <= offset) | |
613 | return (0); | |
614 | length = end - offset; | |
615 | ||
616 | while (length) { | |
617 | start = end; | |
9babb374 | 618 | /* assert(offset <= start) */ |
b128c09f BB |
619 | err = get_next_chunk(dn, &start, offset); |
620 | if (err) | |
621 | return (err); | |
622 | len = trunc ? DMU_OBJECT_END : end - start; | |
623 | ||
624 | tx = dmu_tx_create(os); | |
625 | dmu_tx_hold_free(tx, dn->dn_object, start, len); | |
626 | err = dmu_tx_assign(tx, TXG_WAIT); | |
627 | if (err) { | |
628 | dmu_tx_abort(tx); | |
629 | return (err); | |
630 | } | |
631 | ||
632 | dnode_free_range(dn, start, trunc ? -1 : len, tx); | |
633 | ||
634 | if (start == 0 && free_dnode) { | |
635 | ASSERT(trunc); | |
636 | dnode_free(dn, tx); | |
637 | } | |
638 | ||
639 | length -= end - start; | |
640 | ||
641 | dmu_tx_commit(tx); | |
642 | end = start; | |
643 | } | |
644 | return (0); | |
645 | } | |
646 | ||
647 | int | |
648 | dmu_free_long_range(objset_t *os, uint64_t object, | |
649 | uint64_t offset, uint64_t length) | |
650 | { | |
651 | dnode_t *dn; | |
652 | int err; | |
653 | ||
428870ff | 654 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
655 | if (err != 0) |
656 | return (err); | |
657 | err = dmu_free_long_range_impl(os, dn, offset, length, FALSE); | |
658 | dnode_rele(dn, FTAG); | |
659 | return (err); | |
660 | } | |
661 | ||
662 | int | |
663 | dmu_free_object(objset_t *os, uint64_t object) | |
664 | { | |
665 | dnode_t *dn; | |
666 | dmu_tx_t *tx; | |
667 | int err; | |
668 | ||
428870ff | 669 | err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, |
b128c09f BB |
670 | FTAG, &dn); |
671 | if (err != 0) | |
672 | return (err); | |
673 | if (dn->dn_nlevels == 1) { | |
674 | tx = dmu_tx_create(os); | |
675 | dmu_tx_hold_bonus(tx, object); | |
676 | dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END); | |
677 | err = dmu_tx_assign(tx, TXG_WAIT); | |
678 | if (err == 0) { | |
679 | dnode_free_range(dn, 0, DMU_OBJECT_END, tx); | |
680 | dnode_free(dn, tx); | |
681 | dmu_tx_commit(tx); | |
682 | } else { | |
683 | dmu_tx_abort(tx); | |
684 | } | |
685 | } else { | |
686 | err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE); | |
687 | } | |
688 | dnode_rele(dn, FTAG); | |
689 | return (err); | |
690 | } | |
691 | ||
34dc7c2f BB |
692 | int |
693 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
694 | uint64_t size, dmu_tx_t *tx) | |
695 | { | |
696 | dnode_t *dn; | |
428870ff | 697 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
698 | if (err) |
699 | return (err); | |
700 | ASSERT(offset < UINT64_MAX); | |
701 | ASSERT(size == -1ULL || size <= UINT64_MAX - offset); | |
702 | dnode_free_range(dn, offset, size, tx); | |
703 | dnode_rele(dn, FTAG); | |
704 | return (0); | |
705 | } | |
706 | ||
707 | int | |
708 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
9babb374 | 709 | void *buf, uint32_t flags) |
34dc7c2f BB |
710 | { |
711 | dnode_t *dn; | |
712 | dmu_buf_t **dbp; | |
45d1cae3 | 713 | int numbufs, err; |
34dc7c2f | 714 | |
428870ff | 715 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
716 | if (err) |
717 | return (err); | |
718 | ||
719 | /* | |
720 | * Deal with odd block sizes, where there can't be data past the first | |
721 | * block. If we ever do the tail block optimization, we will need to | |
722 | * handle that here as well. | |
723 | */ | |
45d1cae3 | 724 | if (dn->dn_maxblkid == 0) { |
34dc7c2f BB |
725 | int newsz = offset > dn->dn_datablksz ? 0 : |
726 | MIN(size, dn->dn_datablksz - offset); | |
727 | bzero((char *)buf + newsz, size - newsz); | |
728 | size = newsz; | |
729 | } | |
730 | ||
731 | while (size > 0) { | |
732 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 733 | int i; |
34dc7c2f BB |
734 | |
735 | /* | |
736 | * NB: we could do this block-at-a-time, but it's nice | |
737 | * to be reading in parallel. | |
738 | */ | |
739 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 740 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
741 | if (err) |
742 | break; | |
743 | ||
744 | for (i = 0; i < numbufs; i++) { | |
745 | int tocpy; | |
746 | int bufoff; | |
747 | dmu_buf_t *db = dbp[i]; | |
748 | ||
749 | ASSERT(size > 0); | |
750 | ||
751 | bufoff = offset - db->db_offset; | |
752 | tocpy = (int)MIN(db->db_size - bufoff, size); | |
753 | ||
754 | bcopy((char *)db->db_data + bufoff, buf, tocpy); | |
755 | ||
756 | offset += tocpy; | |
757 | size -= tocpy; | |
758 | buf = (char *)buf + tocpy; | |
759 | } | |
760 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
761 | } | |
762 | dnode_rele(dn, FTAG); | |
763 | return (err); | |
764 | } | |
765 | ||
766 | void | |
767 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
768 | const void *buf, dmu_tx_t *tx) | |
769 | { | |
770 | dmu_buf_t **dbp; | |
771 | int numbufs, i; | |
772 | ||
773 | if (size == 0) | |
774 | return; | |
775 | ||
776 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
777 | FALSE, FTAG, &numbufs, &dbp)); | |
778 | ||
779 | for (i = 0; i < numbufs; i++) { | |
780 | int tocpy; | |
781 | int bufoff; | |
782 | dmu_buf_t *db = dbp[i]; | |
783 | ||
784 | ASSERT(size > 0); | |
785 | ||
786 | bufoff = offset - db->db_offset; | |
787 | tocpy = (int)MIN(db->db_size - bufoff, size); | |
788 | ||
789 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
790 | ||
791 | if (tocpy == db->db_size) | |
792 | dmu_buf_will_fill(db, tx); | |
793 | else | |
794 | dmu_buf_will_dirty(db, tx); | |
795 | ||
60101509 | 796 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
797 | |
798 | if (tocpy == db->db_size) | |
799 | dmu_buf_fill_done(db, tx); | |
800 | ||
801 | offset += tocpy; | |
802 | size -= tocpy; | |
803 | buf = (char *)buf + tocpy; | |
804 | } | |
805 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
806 | } | |
807 | ||
b128c09f BB |
808 | void |
809 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
810 | dmu_tx_t *tx) | |
811 | { | |
812 | dmu_buf_t **dbp; | |
813 | int numbufs, i; | |
814 | ||
815 | if (size == 0) | |
816 | return; | |
817 | ||
818 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
819 | FALSE, FTAG, &numbufs, &dbp)); | |
820 | ||
821 | for (i = 0; i < numbufs; i++) { | |
822 | dmu_buf_t *db = dbp[i]; | |
823 | ||
824 | dmu_buf_will_not_fill(db, tx); | |
825 | } | |
826 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
827 | } | |
828 | ||
428870ff BB |
829 | /* |
830 | * DMU support for xuio | |
831 | */ | |
832 | kstat_t *xuio_ksp = NULL; | |
833 | ||
59e6e7ca BB |
834 | typedef struct xuio_stats { |
835 | /* loaned yet not returned arc_buf */ | |
836 | kstat_named_t xuiostat_onloan_rbuf; | |
837 | kstat_named_t xuiostat_onloan_wbuf; | |
838 | /* whether a copy is made when loaning out a read buffer */ | |
839 | kstat_named_t xuiostat_rbuf_copied; | |
840 | kstat_named_t xuiostat_rbuf_nocopy; | |
841 | /* whether a copy is made when assigning a write buffer */ | |
842 | kstat_named_t xuiostat_wbuf_copied; | |
843 | kstat_named_t xuiostat_wbuf_nocopy; | |
844 | } xuio_stats_t; | |
845 | ||
846 | static xuio_stats_t xuio_stats = { | |
847 | { "onloan_read_buf", KSTAT_DATA_UINT64 }, | |
848 | { "onloan_write_buf", KSTAT_DATA_UINT64 }, | |
849 | { "read_buf_copied", KSTAT_DATA_UINT64 }, | |
850 | { "read_buf_nocopy", KSTAT_DATA_UINT64 }, | |
851 | { "write_buf_copied", KSTAT_DATA_UINT64 }, | |
852 | { "write_buf_nocopy", KSTAT_DATA_UINT64 } | |
853 | }; | |
854 | ||
855 | #define XUIOSTAT_INCR(stat, val) \ | |
856 | atomic_add_64(&xuio_stats.stat.value.ui64, (val)) | |
857 | #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) | |
858 | ||
428870ff BB |
859 | int |
860 | dmu_xuio_init(xuio_t *xuio, int nblk) | |
861 | { | |
862 | dmu_xuio_t *priv; | |
863 | uio_t *uio = &xuio->xu_uio; | |
864 | ||
865 | uio->uio_iovcnt = nblk; | |
866 | uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); | |
867 | ||
868 | priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); | |
869 | priv->cnt = nblk; | |
870 | priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); | |
871 | priv->iovp = uio->uio_iov; | |
872 | XUIO_XUZC_PRIV(xuio) = priv; | |
873 | ||
874 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
875 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); | |
876 | else | |
877 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); | |
878 | ||
879 | return (0); | |
880 | } | |
881 | ||
882 | void | |
883 | dmu_xuio_fini(xuio_t *xuio) | |
884 | { | |
885 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
886 | int nblk = priv->cnt; | |
887 | ||
888 | kmem_free(priv->iovp, nblk * sizeof (iovec_t)); | |
889 | kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); | |
890 | kmem_free(priv, sizeof (dmu_xuio_t)); | |
891 | ||
892 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
893 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); | |
894 | else | |
895 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); | |
896 | } | |
897 | ||
898 | /* | |
899 | * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } | |
900 | * and increase priv->next by 1. | |
901 | */ | |
902 | int | |
903 | dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) | |
904 | { | |
905 | struct iovec *iov; | |
906 | uio_t *uio = &xuio->xu_uio; | |
907 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
908 | int i = priv->next++; | |
909 | ||
910 | ASSERT(i < priv->cnt); | |
911 | ASSERT(off + n <= arc_buf_size(abuf)); | |
912 | iov = uio->uio_iov + i; | |
913 | iov->iov_base = (char *)abuf->b_data + off; | |
914 | iov->iov_len = n; | |
915 | priv->bufs[i] = abuf; | |
916 | return (0); | |
917 | } | |
918 | ||
919 | int | |
920 | dmu_xuio_cnt(xuio_t *xuio) | |
921 | { | |
922 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
923 | return (priv->cnt); | |
924 | } | |
925 | ||
926 | arc_buf_t * | |
927 | dmu_xuio_arcbuf(xuio_t *xuio, int i) | |
928 | { | |
929 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
930 | ||
931 | ASSERT(i < priv->cnt); | |
932 | return (priv->bufs[i]); | |
933 | } | |
934 | ||
935 | void | |
936 | dmu_xuio_clear(xuio_t *xuio, int i) | |
937 | { | |
938 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
939 | ||
940 | ASSERT(i < priv->cnt); | |
941 | priv->bufs[i] = NULL; | |
942 | } | |
943 | ||
944 | static void | |
945 | xuio_stat_init(void) | |
946 | { | |
947 | xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", | |
948 | KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), | |
949 | KSTAT_FLAG_VIRTUAL); | |
950 | if (xuio_ksp != NULL) { | |
951 | xuio_ksp->ks_data = &xuio_stats; | |
952 | kstat_install(xuio_ksp); | |
953 | } | |
954 | } | |
955 | ||
956 | static void | |
957 | xuio_stat_fini(void) | |
958 | { | |
959 | if (xuio_ksp != NULL) { | |
960 | kstat_delete(xuio_ksp); | |
961 | xuio_ksp = NULL; | |
962 | } | |
963 | } | |
964 | ||
965 | void | |
966 | xuio_stat_wbuf_copied() | |
967 | { | |
968 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
969 | } | |
970 | ||
971 | void | |
972 | xuio_stat_wbuf_nocopy() | |
973 | { | |
974 | XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); | |
975 | } | |
976 | ||
34dc7c2f | 977 | #ifdef _KERNEL |
60101509 BB |
978 | |
979 | /* | |
980 | * Copy up to size bytes between arg_buf and req based on the data direction | |
981 | * described by the req. If an entire req's data cannot be transfered the | |
982 | * req's is updated such that it's current index and bv offsets correctly | |
983 | * reference any residual data which could not be copied. The return value | |
984 | * is the number of bytes successfully copied to arg_buf. | |
985 | */ | |
986 | static int | |
987 | dmu_req_copy(void *arg_buf, int size, int *offset, struct request *req) | |
988 | { | |
989 | struct bio_vec *bv; | |
990 | struct req_iterator iter; | |
991 | char *bv_buf; | |
992 | int tocpy; | |
993 | ||
994 | *offset = 0; | |
995 | rq_for_each_segment(bv, req, iter) { | |
996 | ||
997 | /* Fully consumed the passed arg_buf */ | |
998 | ASSERT3S(*offset, <=, size); | |
999 | if (size == *offset) | |
1000 | break; | |
1001 | ||
1002 | /* Skip fully consumed bv's */ | |
1003 | if (bv->bv_len == 0) | |
1004 | continue; | |
1005 | ||
1006 | tocpy = MIN(bv->bv_len, size - *offset); | |
1007 | ASSERT3S(tocpy, >=, 0); | |
1008 | ||
1009 | bv_buf = page_address(bv->bv_page) + bv->bv_offset; | |
1010 | ASSERT3P(bv_buf, !=, NULL); | |
1011 | ||
1012 | if (rq_data_dir(req) == WRITE) | |
1013 | memcpy(arg_buf + *offset, bv_buf, tocpy); | |
1014 | else | |
1015 | memcpy(bv_buf, arg_buf + *offset, tocpy); | |
1016 | ||
1017 | *offset += tocpy; | |
1018 | bv->bv_offset += tocpy; | |
1019 | bv->bv_len -= tocpy; | |
1020 | } | |
1021 | ||
1022 | return 0; | |
1023 | } | |
1024 | ||
34dc7c2f | 1025 | int |
60101509 | 1026 | dmu_read_req(objset_t *os, uint64_t object, struct request *req) |
34dc7c2f | 1027 | { |
60101509 BB |
1028 | uint64_t size = blk_rq_bytes(req); |
1029 | uint64_t offset = blk_rq_pos(req) << 9; | |
34dc7c2f BB |
1030 | dmu_buf_t **dbp; |
1031 | int numbufs, i, err; | |
1032 | ||
1033 | /* | |
1034 | * NB: we could do this block-at-a-time, but it's nice | |
1035 | * to be reading in parallel. | |
1036 | */ | |
60101509 BB |
1037 | err = dmu_buf_hold_array(os, object, offset, size, TRUE, FTAG, |
1038 | &numbufs, &dbp); | |
34dc7c2f BB |
1039 | if (err) |
1040 | return (err); | |
1041 | ||
1042 | for (i = 0; i < numbufs; i++) { | |
60101509 | 1043 | int tocpy, didcpy, bufoff; |
34dc7c2f BB |
1044 | dmu_buf_t *db = dbp[i]; |
1045 | ||
60101509 BB |
1046 | bufoff = offset - db->db_offset; |
1047 | ASSERT3S(bufoff, >=, 0); | |
34dc7c2f | 1048 | |
34dc7c2f | 1049 | tocpy = (int)MIN(db->db_size - bufoff, size); |
60101509 BB |
1050 | if (tocpy == 0) |
1051 | break; | |
34dc7c2f | 1052 | |
60101509 BB |
1053 | err = dmu_req_copy(db->db_data + bufoff, tocpy, &didcpy, req); |
1054 | ||
1055 | if (didcpy < tocpy) | |
1056 | err = EIO; | |
428870ff | 1057 | |
34dc7c2f BB |
1058 | if (err) |
1059 | break; | |
1060 | ||
1061 | size -= tocpy; | |
60101509 BB |
1062 | offset += didcpy; |
1063 | err = 0; | |
34dc7c2f BB |
1064 | } |
1065 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1066 | ||
1067 | return (err); | |
1068 | } | |
1069 | ||
60101509 BB |
1070 | int |
1071 | dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx) | |
34dc7c2f | 1072 | { |
60101509 BB |
1073 | uint64_t size = blk_rq_bytes(req); |
1074 | uint64_t offset = blk_rq_pos(req) << 9; | |
34dc7c2f | 1075 | dmu_buf_t **dbp; |
428870ff | 1076 | int numbufs; |
34dc7c2f | 1077 | int err = 0; |
428870ff | 1078 | int i; |
34dc7c2f | 1079 | |
60101509 BB |
1080 | if (size == 0) |
1081 | return (0); | |
1082 | ||
1083 | err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, | |
1084 | &numbufs, &dbp); | |
34dc7c2f BB |
1085 | if (err) |
1086 | return (err); | |
1087 | ||
1088 | for (i = 0; i < numbufs; i++) { | |
60101509 | 1089 | int tocpy, didcpy, bufoff; |
34dc7c2f BB |
1090 | dmu_buf_t *db = dbp[i]; |
1091 | ||
60101509 BB |
1092 | bufoff = offset - db->db_offset; |
1093 | ASSERT3S(bufoff, >=, 0); | |
34dc7c2f | 1094 | |
34dc7c2f | 1095 | tocpy = (int)MIN(db->db_size - bufoff, size); |
60101509 BB |
1096 | if (tocpy == 0) |
1097 | break; | |
34dc7c2f BB |
1098 | |
1099 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1100 | ||
1101 | if (tocpy == db->db_size) | |
1102 | dmu_buf_will_fill(db, tx); | |
1103 | else | |
1104 | dmu_buf_will_dirty(db, tx); | |
1105 | ||
60101509 | 1106 | err = dmu_req_copy(db->db_data + bufoff, tocpy, &didcpy, req); |
34dc7c2f BB |
1107 | |
1108 | if (tocpy == db->db_size) | |
1109 | dmu_buf_fill_done(db, tx); | |
1110 | ||
60101509 BB |
1111 | if (didcpy < tocpy) |
1112 | err = EIO; | |
1113 | ||
34dc7c2f BB |
1114 | if (err) |
1115 | break; | |
1116 | ||
1117 | size -= tocpy; | |
60101509 BB |
1118 | offset += didcpy; |
1119 | err = 0; | |
34dc7c2f | 1120 | } |
428870ff | 1121 | |
34dc7c2f BB |
1122 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1123 | return (err); | |
1124 | } | |
1125 | ||
872e8d26 BB |
1126 | int |
1127 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
1128 | { | |
1129 | dmu_buf_t **dbp; | |
1130 | int numbufs, i, err; | |
1131 | xuio_t *xuio = NULL; | |
1132 | ||
1133 | /* | |
1134 | * NB: we could do this block-at-a-time, but it's nice | |
1135 | * to be reading in parallel. | |
1136 | */ | |
1137 | err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG, | |
1138 | &numbufs, &dbp); | |
1139 | if (err) | |
1140 | return (err); | |
1141 | ||
1142 | for (i = 0; i < numbufs; i++) { | |
1143 | int tocpy; | |
1144 | int bufoff; | |
1145 | dmu_buf_t *db = dbp[i]; | |
1146 | ||
1147 | ASSERT(size > 0); | |
1148 | ||
1149 | bufoff = uio->uio_loffset - db->db_offset; | |
1150 | tocpy = (int)MIN(db->db_size - bufoff, size); | |
1151 | ||
1152 | if (xuio) { | |
1153 | dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; | |
1154 | arc_buf_t *dbuf_abuf = dbi->db_buf; | |
1155 | arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); | |
1156 | err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); | |
1157 | if (!err) { | |
1158 | uio->uio_resid -= tocpy; | |
1159 | uio->uio_loffset += tocpy; | |
1160 | } | |
1161 | ||
1162 | if (abuf == dbuf_abuf) | |
1163 | XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); | |
1164 | else | |
1165 | XUIOSTAT_BUMP(xuiostat_rbuf_copied); | |
1166 | } else { | |
1167 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1168 | UIO_READ, uio); | |
1169 | } | |
1170 | if (err) | |
1171 | break; | |
1172 | ||
1173 | size -= tocpy; | |
1174 | } | |
1175 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1176 | ||
1177 | return (err); | |
1178 | } | |
1179 | ||
1180 | static int | |
1181 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) | |
1182 | { | |
1183 | dmu_buf_t **dbp; | |
1184 | int numbufs; | |
1185 | int err = 0; | |
1186 | int i; | |
1187 | ||
1188 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1189 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); | |
1190 | if (err) | |
1191 | return (err); | |
1192 | ||
1193 | for (i = 0; i < numbufs; i++) { | |
1194 | int tocpy; | |
1195 | int bufoff; | |
1196 | dmu_buf_t *db = dbp[i]; | |
1197 | ||
1198 | ASSERT(size > 0); | |
1199 | ||
1200 | bufoff = uio->uio_loffset - db->db_offset; | |
1201 | tocpy = (int)MIN(db->db_size - bufoff, size); | |
1202 | ||
1203 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1204 | ||
1205 | if (tocpy == db->db_size) | |
1206 | dmu_buf_will_fill(db, tx); | |
1207 | else | |
1208 | dmu_buf_will_dirty(db, tx); | |
1209 | ||
1210 | /* | |
1211 | * XXX uiomove could block forever (eg.nfs-backed | |
1212 | * pages). There needs to be a uiolockdown() function | |
1213 | * to lock the pages in memory, so that uiomove won't | |
1214 | * block. | |
1215 | */ | |
1216 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1217 | UIO_WRITE, uio); | |
1218 | ||
1219 | if (tocpy == db->db_size) | |
1220 | dmu_buf_fill_done(db, tx); | |
1221 | ||
1222 | if (err) | |
1223 | break; | |
1224 | ||
1225 | size -= tocpy; | |
1226 | } | |
1227 | ||
1228 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1229 | return (err); | |
1230 | } | |
1231 | ||
428870ff BB |
1232 | int |
1233 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1234 | dmu_tx_t *tx) | |
1235 | { | |
572e2857 BB |
1236 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1237 | dnode_t *dn; | |
1238 | int err; | |
1239 | ||
428870ff BB |
1240 | if (size == 0) |
1241 | return (0); | |
1242 | ||
572e2857 BB |
1243 | DB_DNODE_ENTER(db); |
1244 | dn = DB_DNODE(db); | |
1245 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1246 | DB_DNODE_EXIT(db); | |
1247 | ||
1248 | return (err); | |
428870ff BB |
1249 | } |
1250 | ||
1251 | int | |
1252 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1253 | dmu_tx_t *tx) | |
1254 | { | |
1255 | dnode_t *dn; | |
1256 | int err; | |
1257 | ||
1258 | if (size == 0) | |
1259 | return (0); | |
1260 | ||
1261 | err = dnode_hold(os, object, FTAG, &dn); | |
1262 | if (err) | |
1263 | return (err); | |
1264 | ||
1265 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1266 | ||
1267 | dnode_rele(dn, FTAG); | |
1268 | ||
1269 | return (err); | |
1270 | } | |
872e8d26 | 1271 | #endif /* _KERNEL */ |
34dc7c2f | 1272 | |
9babb374 BB |
1273 | /* |
1274 | * Allocate a loaned anonymous arc buffer. | |
1275 | */ | |
1276 | arc_buf_t * | |
1277 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1278 | { | |
572e2857 BB |
1279 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
1280 | spa_t *spa; | |
9babb374 | 1281 | |
572e2857 BB |
1282 | DB_GET_SPA(&spa, db); |
1283 | return (arc_loan_buf(spa, size)); | |
9babb374 BB |
1284 | } |
1285 | ||
1286 | /* | |
1287 | * Free a loaned arc buffer. | |
1288 | */ | |
1289 | void | |
1290 | dmu_return_arcbuf(arc_buf_t *buf) | |
1291 | { | |
1292 | arc_return_buf(buf, FTAG); | |
1293 | VERIFY(arc_buf_remove_ref(buf, FTAG) == 1); | |
1294 | } | |
1295 | ||
1296 | /* | |
1297 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1298 | * If this is not possible copy the contents of passed arc buf via | |
1299 | * dmu_write(). | |
1300 | */ | |
1301 | void | |
1302 | dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, | |
1303 | dmu_tx_t *tx) | |
1304 | { | |
572e2857 BB |
1305 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1306 | dnode_t *dn; | |
9babb374 BB |
1307 | dmu_buf_impl_t *db; |
1308 | uint32_t blksz = (uint32_t)arc_buf_size(buf); | |
1309 | uint64_t blkid; | |
1310 | ||
572e2857 BB |
1311 | DB_DNODE_ENTER(dbuf); |
1312 | dn = DB_DNODE(dbuf); | |
9babb374 BB |
1313 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
1314 | blkid = dbuf_whichblock(dn, offset); | |
1315 | VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); | |
1316 | rw_exit(&dn->dn_struct_rwlock); | |
572e2857 | 1317 | DB_DNODE_EXIT(dbuf); |
9babb374 BB |
1318 | |
1319 | if (offset == db->db.db_offset && blksz == db->db.db_size) { | |
1320 | dbuf_assign_arcbuf(db, buf, tx); | |
1321 | dbuf_rele(db, FTAG); | |
1322 | } else { | |
572e2857 BB |
1323 | objset_t *os; |
1324 | uint64_t object; | |
1325 | ||
1326 | DB_DNODE_ENTER(dbuf); | |
1327 | dn = DB_DNODE(dbuf); | |
1328 | os = dn->dn_objset; | |
1329 | object = dn->dn_object; | |
1330 | DB_DNODE_EXIT(dbuf); | |
1331 | ||
9babb374 | 1332 | dbuf_rele(db, FTAG); |
572e2857 | 1333 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 | 1334 | dmu_return_arcbuf(buf); |
428870ff | 1335 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
9babb374 BB |
1336 | } |
1337 | } | |
1338 | ||
34dc7c2f | 1339 | typedef struct { |
428870ff BB |
1340 | dbuf_dirty_record_t *dsa_dr; |
1341 | dmu_sync_cb_t *dsa_done; | |
1342 | zgd_t *dsa_zgd; | |
1343 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1344 | } dmu_sync_arg_t; |
1345 | ||
b128c09f BB |
1346 | /* ARGSUSED */ |
1347 | static void | |
1348 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1349 | { | |
428870ff BB |
1350 | dmu_sync_arg_t *dsa = varg; |
1351 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1352 | blkptr_t *bp = zio->io_bp; |
1353 | ||
428870ff BB |
1354 | if (zio->io_error == 0) { |
1355 | if (BP_IS_HOLE(bp)) { | |
1356 | /* | |
1357 | * A block of zeros may compress to a hole, but the | |
1358 | * block size still needs to be known for replay. | |
1359 | */ | |
1360 | BP_SET_LSIZE(bp, db->db_size); | |
1361 | } else { | |
428870ff BB |
1362 | ASSERT(BP_GET_LEVEL(bp) == 0); |
1363 | bp->blk_fill = 1; | |
1364 | } | |
b128c09f BB |
1365 | } |
1366 | } | |
1367 | ||
428870ff BB |
1368 | static void |
1369 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1370 | { | |
1371 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1372 | } | |
1373 | ||
34dc7c2f BB |
1374 | /* ARGSUSED */ |
1375 | static void | |
1376 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1377 | { | |
428870ff BB |
1378 | dmu_sync_arg_t *dsa = varg; |
1379 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1380 | dmu_buf_impl_t *db = dr->dr_dbuf; |
34dc7c2f | 1381 | |
34dc7c2f BB |
1382 | mutex_enter(&db->db_mtx); |
1383 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff BB |
1384 | if (zio->io_error == 0) { |
1385 | dr->dt.dl.dr_overridden_by = *zio->io_bp; | |
1386 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1387 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
1388 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by)) | |
1389 | BP_ZERO(&dr->dt.dl.dr_overridden_by); | |
1390 | } else { | |
1391 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1392 | } | |
34dc7c2f BB |
1393 | cv_broadcast(&db->db_changed); |
1394 | mutex_exit(&db->db_mtx); | |
1395 | ||
428870ff | 1396 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1397 | |
428870ff BB |
1398 | kmem_free(dsa, sizeof (*dsa)); |
1399 | } | |
1400 | ||
1401 | static void | |
1402 | dmu_sync_late_arrival_done(zio_t *zio) | |
1403 | { | |
1404 | blkptr_t *bp = zio->io_bp; | |
1405 | dmu_sync_arg_t *dsa = zio->io_private; | |
1406 | ||
1407 | if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { | |
1408 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1409 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1410 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1411 | } | |
1412 | ||
1413 | dmu_tx_commit(dsa->dsa_tx); | |
1414 | ||
1415 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1416 | ||
1417 | kmem_free(dsa, sizeof (*dsa)); | |
1418 | } | |
1419 | ||
1420 | static int | |
1421 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
1422 | zio_prop_t *zp, zbookmark_t *zb) | |
1423 | { | |
1424 | dmu_sync_arg_t *dsa; | |
1425 | dmu_tx_t *tx; | |
1426 | ||
1427 | tx = dmu_tx_create(os); | |
1428 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1429 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1430 | dmu_tx_abort(tx); | |
1431 | return (EIO); /* Make zl_get_data do txg_waited_synced() */ | |
1432 | } | |
1433 | ||
1434 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); | |
1435 | dsa->dsa_dr = NULL; | |
1436 | dsa->dsa_done = done; | |
1437 | dsa->dsa_zgd = zgd; | |
1438 | dsa->dsa_tx = tx; | |
1439 | ||
1440 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, | |
1441 | zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, | |
1442 | dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa, | |
1443 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); | |
1444 | ||
1445 | return (0); | |
34dc7c2f BB |
1446 | } |
1447 | ||
1448 | /* | |
1449 | * Intent log support: sync the block associated with db to disk. | |
1450 | * N.B. and XXX: the caller is responsible for making sure that the | |
1451 | * data isn't changing while dmu_sync() is writing it. | |
1452 | * | |
1453 | * Return values: | |
1454 | * | |
1455 | * EEXIST: this txg has already been synced, so there's nothing to to. | |
1456 | * The caller should not log the write. | |
1457 | * | |
1458 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1459 | * The caller should not log the write. | |
1460 | * | |
1461 | * EALREADY: this block is already in the process of being synced. | |
1462 | * The caller should track its progress (somehow). | |
1463 | * | |
428870ff BB |
1464 | * EIO: could not do the I/O. |
1465 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1466 | * |
428870ff BB |
1467 | * 0: the I/O has been initiated. |
1468 | * The caller should log this blkptr in the done callback. | |
1469 | * It is possible that the I/O will fail, in which case | |
1470 | * the error will be reported to the done callback and | |
1471 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1472 | */ |
1473 | int | |
428870ff | 1474 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1475 | { |
428870ff BB |
1476 | blkptr_t *bp = zgd->zgd_bp; |
1477 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; | |
1478 | objset_t *os = db->db_objset; | |
1479 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
34dc7c2f | 1480 | dbuf_dirty_record_t *dr; |
428870ff | 1481 | dmu_sync_arg_t *dsa; |
34dc7c2f | 1482 | zbookmark_t zb; |
428870ff | 1483 | zio_prop_t zp; |
572e2857 | 1484 | dnode_t *dn; |
34dc7c2f | 1485 | |
428870ff | 1486 | ASSERT(pio != NULL); |
34dc7c2f BB |
1487 | ASSERT(BP_IS_HOLE(bp)); |
1488 | ASSERT(txg != 0); | |
1489 | ||
428870ff BB |
1490 | SET_BOOKMARK(&zb, ds->ds_object, |
1491 | db->db.db_object, db->db_level, db->db_blkid); | |
1492 | ||
572e2857 BB |
1493 | DB_DNODE_ENTER(db); |
1494 | dn = DB_DNODE(db); | |
1495 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); | |
1496 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
1497 | |
1498 | /* | |
428870ff | 1499 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1500 | */ |
428870ff BB |
1501 | if (txg > spa_freeze_txg(os->os_spa)) |
1502 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1503 | |
1504 | /* | |
428870ff BB |
1505 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1506 | * and us. If we determine that this txg is not yet syncing, | |
1507 | * but it begins to sync a moment later, that's OK because the | |
1508 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1509 | */ |
428870ff BB |
1510 | mutex_enter(&db->db_mtx); |
1511 | ||
1512 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1513 | /* |
428870ff | 1514 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1515 | */ |
428870ff | 1516 | mutex_exit(&db->db_mtx); |
34dc7c2f BB |
1517 | return (EEXIST); |
1518 | } | |
1519 | ||
428870ff BB |
1520 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1521 | /* | |
1522 | * This txg is currently syncing, so we can't mess with | |
1523 | * the dirty record anymore; just write a new log block. | |
1524 | */ | |
1525 | mutex_exit(&db->db_mtx); | |
1526 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1527 | } |
1528 | ||
1529 | dr = db->db_last_dirty; | |
428870ff | 1530 | while (dr && dr->dr_txg != txg) |
34dc7c2f | 1531 | dr = dr->dr_next; |
428870ff BB |
1532 | |
1533 | if (dr == NULL) { | |
34dc7c2f | 1534 | /* |
428870ff | 1535 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1536 | * There's no need to log writes to freed blocks, so we're done. |
1537 | */ | |
1538 | mutex_exit(&db->db_mtx); | |
34dc7c2f BB |
1539 | return (ENOENT); |
1540 | } | |
1541 | ||
1542 | ASSERT(dr->dr_txg == txg); | |
428870ff BB |
1543 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1544 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1545 | /* |
428870ff BB |
1546 | * We have already issued a sync write for this buffer, |
1547 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1548 | * have been dirtied since, or we would have cleared the state. |
1549 | */ | |
34dc7c2f | 1550 | mutex_exit(&db->db_mtx); |
428870ff | 1551 | return (EALREADY); |
34dc7c2f BB |
1552 | } |
1553 | ||
428870ff | 1554 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1555 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1556 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1557 | |
428870ff BB |
1558 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); |
1559 | dsa->dsa_dr = dr; | |
1560 | dsa->dsa_done = done; | |
1561 | dsa->dsa_zgd = zgd; | |
1562 | dsa->dsa_tx = NULL; | |
b128c09f | 1563 | |
428870ff BB |
1564 | zio_nowait(arc_write(pio, os->os_spa, txg, |
1565 | bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp, | |
1566 | dmu_sync_ready, dmu_sync_done, dsa, | |
1567 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); | |
b128c09f | 1568 | |
428870ff | 1569 | return (0); |
34dc7c2f BB |
1570 | } |
1571 | ||
1572 | int | |
1573 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
1574 | dmu_tx_t *tx) | |
1575 | { | |
1576 | dnode_t *dn; | |
1577 | int err; | |
1578 | ||
428870ff | 1579 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1580 | if (err) |
1581 | return (err); | |
1582 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1583 | dnode_rele(dn, FTAG); | |
1584 | return (err); | |
1585 | } | |
1586 | ||
1587 | void | |
1588 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
1589 | dmu_tx_t *tx) | |
1590 | { | |
1591 | dnode_t *dn; | |
1592 | ||
1593 | /* XXX assumes dnode_hold will not get an i/o error */ | |
428870ff | 1594 | (void) dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1595 | ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS); |
1596 | dn->dn_checksum = checksum; | |
1597 | dnode_setdirty(dn, tx); | |
1598 | dnode_rele(dn, FTAG); | |
1599 | } | |
1600 | ||
1601 | void | |
1602 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
1603 | dmu_tx_t *tx) | |
1604 | { | |
1605 | dnode_t *dn; | |
1606 | ||
1607 | /* XXX assumes dnode_hold will not get an i/o error */ | |
428870ff | 1608 | (void) dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1609 | ASSERT(compress < ZIO_COMPRESS_FUNCTIONS); |
1610 | dn->dn_compress = compress; | |
1611 | dnode_setdirty(dn, tx); | |
1612 | dnode_rele(dn, FTAG); | |
1613 | } | |
1614 | ||
428870ff BB |
1615 | int zfs_mdcomp_disable = 0; |
1616 | ||
1617 | void | |
1618 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) | |
1619 | { | |
1620 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
572e2857 BB |
1621 | boolean_t ismd = (level > 0 || dmu_ot[type].ot_metadata || |
1622 | (wp & WP_SPILL)); | |
428870ff BB |
1623 | enum zio_checksum checksum = os->os_checksum; |
1624 | enum zio_compress compress = os->os_compress; | |
1625 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; | |
1626 | boolean_t dedup; | |
1627 | boolean_t dedup_verify = os->os_dedup_verify; | |
1628 | int copies = os->os_copies; | |
1629 | ||
1630 | /* | |
1631 | * Determine checksum setting. | |
1632 | */ | |
1633 | if (ismd) { | |
1634 | /* | |
1635 | * Metadata always gets checksummed. If the data | |
1636 | * checksum is multi-bit correctable, and it's not a | |
1637 | * ZBT-style checksum, then it's suitable for metadata | |
1638 | * as well. Otherwise, the metadata checksum defaults | |
1639 | * to fletcher4. | |
1640 | */ | |
1641 | if (zio_checksum_table[checksum].ci_correctable < 1 || | |
1642 | zio_checksum_table[checksum].ci_eck) | |
1643 | checksum = ZIO_CHECKSUM_FLETCHER_4; | |
1644 | } else { | |
1645 | checksum = zio_checksum_select(dn->dn_checksum, checksum); | |
1646 | } | |
1647 | ||
1648 | /* | |
1649 | * Determine compression setting. | |
1650 | */ | |
1651 | if (ismd) { | |
1652 | /* | |
1653 | * XXX -- we should design a compression algorithm | |
1654 | * that specializes in arrays of bps. | |
1655 | */ | |
1656 | compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY : | |
1657 | ZIO_COMPRESS_LZJB; | |
1658 | } else { | |
1659 | compress = zio_compress_select(dn->dn_compress, compress); | |
1660 | } | |
1661 | ||
1662 | /* | |
1663 | * Determine dedup setting. If we are in dmu_sync(), we won't | |
1664 | * actually dedup now because that's all done in syncing context; | |
1665 | * but we do want to use the dedup checkum. If the checksum is not | |
1666 | * strong enough to ensure unique signatures, force dedup_verify. | |
1667 | */ | |
1668 | dedup = (!ismd && dedup_checksum != ZIO_CHECKSUM_OFF); | |
1669 | if (dedup) { | |
1670 | checksum = dedup_checksum; | |
1671 | if (!zio_checksum_table[checksum].ci_dedup) | |
1672 | dedup_verify = 1; | |
1673 | } | |
1674 | ||
1675 | if (wp & WP_DMU_SYNC) | |
1676 | dedup = 0; | |
1677 | ||
1678 | if (wp & WP_NOFILL) { | |
1679 | ASSERT(!ismd && level == 0); | |
1680 | checksum = ZIO_CHECKSUM_OFF; | |
1681 | compress = ZIO_COMPRESS_OFF; | |
1682 | dedup = B_FALSE; | |
1683 | } | |
1684 | ||
1685 | zp->zp_checksum = checksum; | |
1686 | zp->zp_compress = compress; | |
1687 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; | |
1688 | zp->zp_level = level; | |
1689 | zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa)); | |
1690 | zp->zp_dedup = dedup; | |
1691 | zp->zp_dedup_verify = dedup && dedup_verify; | |
1692 | } | |
1693 | ||
34dc7c2f BB |
1694 | int |
1695 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
1696 | { | |
1697 | dnode_t *dn; | |
1698 | int i, err; | |
1699 | ||
428870ff | 1700 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1701 | if (err) |
1702 | return (err); | |
1703 | /* | |
1704 | * Sync any current changes before | |
1705 | * we go trundling through the block pointers. | |
1706 | */ | |
1707 | for (i = 0; i < TXG_SIZE; i++) { | |
1708 | if (list_link_active(&dn->dn_dirty_link[i])) | |
1709 | break; | |
1710 | } | |
1711 | if (i != TXG_SIZE) { | |
1712 | dnode_rele(dn, FTAG); | |
1713 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 1714 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1715 | if (err) |
1716 | return (err); | |
1717 | } | |
1718 | ||
b128c09f | 1719 | err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); |
34dc7c2f BB |
1720 | dnode_rele(dn, FTAG); |
1721 | ||
1722 | return (err); | |
1723 | } | |
1724 | ||
1725 | void | |
1726 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
1727 | { | |
428870ff | 1728 | dnode_phys_t *dnp; |
d6320ddb | 1729 | int i; |
428870ff | 1730 | |
34dc7c2f BB |
1731 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
1732 | mutex_enter(&dn->dn_mtx); | |
1733 | ||
428870ff BB |
1734 | dnp = dn->dn_phys; |
1735 | ||
34dc7c2f BB |
1736 | doi->doi_data_block_size = dn->dn_datablksz; |
1737 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
1738 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
1739 | doi->doi_type = dn->dn_type; |
1740 | doi->doi_bonus_type = dn->dn_bonustype; | |
1741 | doi->doi_bonus_size = dn->dn_bonuslen; | |
34dc7c2f BB |
1742 | doi->doi_indirection = dn->dn_nlevels; |
1743 | doi->doi_checksum = dn->dn_checksum; | |
1744 | doi->doi_compress = dn->dn_compress; | |
428870ff BB |
1745 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
1746 | doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz; | |
1747 | doi->doi_fill_count = 0; | |
d6320ddb | 1748 | for (i = 0; i < dnp->dn_nblkptr; i++) |
428870ff | 1749 | doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill; |
34dc7c2f BB |
1750 | |
1751 | mutex_exit(&dn->dn_mtx); | |
1752 | rw_exit(&dn->dn_struct_rwlock); | |
1753 | } | |
1754 | ||
1755 | /* | |
1756 | * Get information on a DMU object. | |
1757 | * If doi is NULL, just indicates whether the object exists. | |
1758 | */ | |
1759 | int | |
1760 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
1761 | { | |
1762 | dnode_t *dn; | |
428870ff | 1763 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1764 | |
1765 | if (err) | |
1766 | return (err); | |
1767 | ||
1768 | if (doi != NULL) | |
1769 | dmu_object_info_from_dnode(dn, doi); | |
1770 | ||
1771 | dnode_rele(dn, FTAG); | |
1772 | return (0); | |
1773 | } | |
1774 | ||
1775 | /* | |
1776 | * As above, but faster; can be used when you have a held dbuf in hand. | |
1777 | */ | |
1778 | void | |
572e2857 | 1779 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 1780 | { |
572e2857 BB |
1781 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
1782 | ||
1783 | DB_DNODE_ENTER(db); | |
1784 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
1785 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
1786 | } |
1787 | ||
1788 | /* | |
1789 | * Faster still when you only care about the size. | |
1790 | * This is specifically optimized for zfs_getattr(). | |
1791 | */ | |
1792 | void | |
572e2857 BB |
1793 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
1794 | u_longlong_t *nblk512) | |
34dc7c2f | 1795 | { |
572e2857 BB |
1796 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
1797 | dnode_t *dn; | |
1798 | ||
1799 | DB_DNODE_ENTER(db); | |
1800 | dn = DB_DNODE(db); | |
34dc7c2f BB |
1801 | |
1802 | *blksize = dn->dn_datablksz; | |
1803 | /* add 1 for dnode space */ | |
1804 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> | |
1805 | SPA_MINBLOCKSHIFT) + 1; | |
572e2857 | 1806 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1807 | } |
1808 | ||
1809 | void | |
1810 | byteswap_uint64_array(void *vbuf, size_t size) | |
1811 | { | |
1812 | uint64_t *buf = vbuf; | |
1813 | size_t count = size >> 3; | |
1814 | int i; | |
1815 | ||
1816 | ASSERT((size & 7) == 0); | |
1817 | ||
1818 | for (i = 0; i < count; i++) | |
1819 | buf[i] = BSWAP_64(buf[i]); | |
1820 | } | |
1821 | ||
1822 | void | |
1823 | byteswap_uint32_array(void *vbuf, size_t size) | |
1824 | { | |
1825 | uint32_t *buf = vbuf; | |
1826 | size_t count = size >> 2; | |
1827 | int i; | |
1828 | ||
1829 | ASSERT((size & 3) == 0); | |
1830 | ||
1831 | for (i = 0; i < count; i++) | |
1832 | buf[i] = BSWAP_32(buf[i]); | |
1833 | } | |
1834 | ||
1835 | void | |
1836 | byteswap_uint16_array(void *vbuf, size_t size) | |
1837 | { | |
1838 | uint16_t *buf = vbuf; | |
1839 | size_t count = size >> 1; | |
1840 | int i; | |
1841 | ||
1842 | ASSERT((size & 1) == 0); | |
1843 | ||
1844 | for (i = 0; i < count; i++) | |
1845 | buf[i] = BSWAP_16(buf[i]); | |
1846 | } | |
1847 | ||
1848 | /* ARGSUSED */ | |
1849 | void | |
1850 | byteswap_uint8_array(void *vbuf, size_t size) | |
1851 | { | |
1852 | } | |
1853 | ||
1854 | void | |
1855 | dmu_init(void) | |
1856 | { | |
428870ff | 1857 | zfs_dbgmsg_init(); |
572e2857 BB |
1858 | sa_cache_init(); |
1859 | xuio_stat_init(); | |
1860 | dmu_objset_init(); | |
34dc7c2f | 1861 | dnode_init(); |
572e2857 | 1862 | dbuf_init(); |
428870ff | 1863 | zfetch_init(); |
34dc7c2f BB |
1864 | arc_init(); |
1865 | l2arc_init(); | |
1866 | } | |
1867 | ||
1868 | void | |
1869 | dmu_fini(void) | |
1870 | { | |
572e2857 | 1871 | l2arc_fini(); |
34dc7c2f | 1872 | arc_fini(); |
428870ff | 1873 | zfetch_fini(); |
34dc7c2f | 1874 | dbuf_fini(); |
572e2857 BB |
1875 | dnode_fini(); |
1876 | dmu_objset_fini(); | |
428870ff BB |
1877 | xuio_stat_fini(); |
1878 | sa_cache_fini(); | |
1879 | zfs_dbgmsg_fini(); | |
34dc7c2f | 1880 | } |
c28b2279 BB |
1881 | |
1882 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
1883 | EXPORT_SYMBOL(dmu_bonus_hold); | |
1884 | EXPORT_SYMBOL(dmu_free_range); | |
1885 | EXPORT_SYMBOL(dmu_read); | |
1886 | EXPORT_SYMBOL(dmu_write); | |
1887 | ||
1888 | /* Get information on a DMU object. */ | |
1889 | EXPORT_SYMBOL(dmu_object_info); | |
1890 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
1891 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
1892 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
1893 | ||
1894 | EXPORT_SYMBOL(dmu_object_set_blocksize); | |
1895 | EXPORT_SYMBOL(dmu_object_set_checksum); | |
1896 | EXPORT_SYMBOL(dmu_object_set_compress); | |
1897 | ||
1898 | EXPORT_SYMBOL(dmu_ot); | |
1899 | #endif |