]>
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. |
9ae529ec | 23 | * Copyright (c) 2012 by Delphix. All rights reserved. |
3a17a7a9 | 24 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
34dc7c2f BB |
25 | */ |
26 | ||
34dc7c2f BB |
27 | #include <sys/dmu.h> |
28 | #include <sys/dmu_impl.h> | |
29 | #include <sys/dmu_tx.h> | |
30 | #include <sys/dbuf.h> | |
31 | #include <sys/dnode.h> | |
32 | #include <sys/zfs_context.h> | |
33 | #include <sys/dmu_objset.h> | |
34 | #include <sys/dmu_traverse.h> | |
35 | #include <sys/dsl_dataset.h> | |
36 | #include <sys/dsl_dir.h> | |
37 | #include <sys/dsl_pool.h> | |
38 | #include <sys/dsl_synctask.h> | |
39 | #include <sys/dsl_prop.h> | |
40 | #include <sys/dmu_zfetch.h> | |
41 | #include <sys/zfs_ioctl.h> | |
42 | #include <sys/zap.h> | |
43 | #include <sys/zio_checksum.h> | |
428870ff | 44 | #include <sys/sa.h> |
34dc7c2f BB |
45 | #ifdef _KERNEL |
46 | #include <sys/vmsystm.h> | |
b128c09f | 47 | #include <sys/zfs_znode.h> |
34dc7c2f BB |
48 | #endif |
49 | ||
50 | const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { | |
9ae529ec CS |
51 | { DMU_BSWAP_UINT8, TRUE, "unallocated" }, |
52 | { DMU_BSWAP_ZAP, TRUE, "object directory" }, | |
53 | { DMU_BSWAP_UINT64, TRUE, "object array" }, | |
54 | { DMU_BSWAP_UINT8, TRUE, "packed nvlist" }, | |
55 | { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" }, | |
56 | { DMU_BSWAP_UINT64, TRUE, "bpobj" }, | |
57 | { DMU_BSWAP_UINT64, TRUE, "bpobj header" }, | |
58 | { DMU_BSWAP_UINT64, TRUE, "SPA space map header" }, | |
59 | { DMU_BSWAP_UINT64, TRUE, "SPA space map" }, | |
60 | { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" }, | |
61 | { DMU_BSWAP_DNODE, TRUE, "DMU dnode" }, | |
62 | { DMU_BSWAP_OBJSET, TRUE, "DMU objset" }, | |
63 | { DMU_BSWAP_UINT64, TRUE, "DSL directory" }, | |
64 | { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"}, | |
65 | { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" }, | |
66 | { DMU_BSWAP_ZAP, TRUE, "DSL props" }, | |
67 | { DMU_BSWAP_UINT64, TRUE, "DSL dataset" }, | |
68 | { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" }, | |
69 | { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" }, | |
70 | { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" }, | |
71 | { DMU_BSWAP_ZAP, TRUE, "ZFS directory" }, | |
72 | { DMU_BSWAP_ZAP, TRUE, "ZFS master node" }, | |
73 | { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" }, | |
74 | { DMU_BSWAP_UINT8, FALSE, "zvol object" }, | |
75 | { DMU_BSWAP_ZAP, TRUE, "zvol prop" }, | |
76 | { DMU_BSWAP_UINT8, FALSE, "other uint8[]" }, | |
77 | { DMU_BSWAP_UINT64, FALSE, "other uint64[]" }, | |
78 | { DMU_BSWAP_ZAP, TRUE, "other ZAP" }, | |
79 | { DMU_BSWAP_ZAP, TRUE, "persistent error log" }, | |
80 | { DMU_BSWAP_UINT8, TRUE, "SPA history" }, | |
81 | { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" }, | |
82 | { DMU_BSWAP_ZAP, TRUE, "Pool properties" }, | |
83 | { DMU_BSWAP_ZAP, TRUE, "DSL permissions" }, | |
84 | { DMU_BSWAP_ACL, TRUE, "ZFS ACL" }, | |
85 | { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" }, | |
86 | { DMU_BSWAP_UINT8, TRUE, "FUID table" }, | |
87 | { DMU_BSWAP_UINT64, TRUE, "FUID table size" }, | |
88 | { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"}, | |
89 | { DMU_BSWAP_ZAP, TRUE, "scan work queue" }, | |
90 | { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" }, | |
91 | { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" }, | |
92 | { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"}, | |
93 | { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" }, | |
94 | { DMU_BSWAP_ZAP, TRUE, "DDT statistics" }, | |
95 | { DMU_BSWAP_UINT8, TRUE, "System attributes" }, | |
96 | { DMU_BSWAP_ZAP, TRUE, "SA master node" }, | |
97 | { DMU_BSWAP_ZAP, TRUE, "SA attr registration" }, | |
98 | { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" }, | |
99 | { DMU_BSWAP_ZAP, TRUE, "scan translations" }, | |
100 | { DMU_BSWAP_UINT8, FALSE, "deduplicated block" }, | |
101 | { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" }, | |
102 | { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" }, | |
103 | { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" }, | |
104 | { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" } | |
105 | }; | |
106 | ||
107 | const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { | |
108 | { byteswap_uint8_array, "uint8" }, | |
109 | { byteswap_uint16_array, "uint16" }, | |
110 | { byteswap_uint32_array, "uint32" }, | |
111 | { byteswap_uint64_array, "uint64" }, | |
112 | { zap_byteswap, "zap" }, | |
113 | { dnode_buf_byteswap, "dnode" }, | |
114 | { dmu_objset_byteswap, "objset" }, | |
115 | { zfs_znode_byteswap, "znode" }, | |
116 | { zfs_oldacl_byteswap, "oldacl" }, | |
117 | { zfs_acl_byteswap, "acl" } | |
34dc7c2f BB |
118 | }; |
119 | ||
120 | int | |
121 | dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, | |
428870ff | 122 | void *tag, dmu_buf_t **dbp, int flags) |
34dc7c2f BB |
123 | { |
124 | dnode_t *dn; | |
125 | uint64_t blkid; | |
126 | dmu_buf_impl_t *db; | |
127 | int err; | |
428870ff | 128 | int db_flags = DB_RF_CANFAIL; |
34dc7c2f | 129 | |
428870ff BB |
130 | if (flags & DMU_READ_NO_PREFETCH) |
131 | db_flags |= DB_RF_NOPREFETCH; | |
132 | ||
133 | err = dnode_hold(os, object, FTAG, &dn); | |
34dc7c2f BB |
134 | if (err) |
135 | return (err); | |
136 | blkid = dbuf_whichblock(dn, offset); | |
137 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
138 | db = dbuf_hold(dn, blkid, tag); | |
139 | rw_exit(&dn->dn_struct_rwlock); | |
140 | if (db == NULL) { | |
141 | err = EIO; | |
142 | } else { | |
428870ff | 143 | err = dbuf_read(db, NULL, db_flags); |
34dc7c2f BB |
144 | if (err) { |
145 | dbuf_rele(db, tag); | |
146 | db = NULL; | |
147 | } | |
148 | } | |
149 | ||
150 | dnode_rele(dn, FTAG); | |
572e2857 | 151 | *dbp = &db->db; /* NULL db plus first field offset is NULL */ |
34dc7c2f BB |
152 | return (err); |
153 | } | |
154 | ||
155 | int | |
156 | dmu_bonus_max(void) | |
157 | { | |
158 | return (DN_MAX_BONUSLEN); | |
159 | } | |
160 | ||
161 | int | |
572e2857 | 162 | dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) |
34dc7c2f | 163 | { |
572e2857 BB |
164 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
165 | dnode_t *dn; | |
166 | int error; | |
34dc7c2f | 167 | |
572e2857 BB |
168 | DB_DNODE_ENTER(db); |
169 | dn = DB_DNODE(db); | |
170 | ||
171 | if (dn->dn_bonus != db) { | |
172 | error = EINVAL; | |
173 | } else if (newsize < 0 || newsize > db_fake->db_size) { | |
174 | error = EINVAL; | |
175 | } else { | |
176 | dnode_setbonuslen(dn, newsize, tx); | |
177 | error = 0; | |
178 | } | |
179 | ||
180 | DB_DNODE_EXIT(db); | |
181 | return (error); | |
34dc7c2f BB |
182 | } |
183 | ||
428870ff | 184 | int |
572e2857 | 185 | dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) |
428870ff | 186 | { |
572e2857 BB |
187 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
188 | dnode_t *dn; | |
189 | int error; | |
428870ff | 190 | |
572e2857 BB |
191 | DB_DNODE_ENTER(db); |
192 | dn = DB_DNODE(db); | |
428870ff | 193 | |
9ae529ec | 194 | if (!DMU_OT_IS_VALID(type)) { |
572e2857 BB |
195 | error = EINVAL; |
196 | } else if (dn->dn_bonus != db) { | |
197 | error = EINVAL; | |
198 | } else { | |
199 | dnode_setbonus_type(dn, type, tx); | |
200 | error = 0; | |
201 | } | |
428870ff | 202 | |
572e2857 BB |
203 | DB_DNODE_EXIT(db); |
204 | return (error); | |
205 | } | |
206 | ||
207 | dmu_object_type_t | |
208 | dmu_get_bonustype(dmu_buf_t *db_fake) | |
209 | { | |
210 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; | |
211 | dnode_t *dn; | |
212 | dmu_object_type_t type; | |
213 | ||
214 | DB_DNODE_ENTER(db); | |
215 | dn = DB_DNODE(db); | |
216 | type = dn->dn_bonustype; | |
217 | DB_DNODE_EXIT(db); | |
218 | ||
219 | return (type); | |
428870ff BB |
220 | } |
221 | ||
222 | int | |
223 | dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) | |
224 | { | |
225 | dnode_t *dn; | |
226 | int error; | |
227 | ||
228 | error = dnode_hold(os, object, FTAG, &dn); | |
229 | dbuf_rm_spill(dn, tx); | |
230 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
231 | dnode_rm_spill(dn, tx); | |
232 | rw_exit(&dn->dn_struct_rwlock); | |
233 | dnode_rele(dn, FTAG); | |
234 | return (error); | |
235 | } | |
236 | ||
34dc7c2f BB |
237 | /* |
238 | * returns ENOENT, EIO, or 0. | |
239 | */ | |
240 | int | |
241 | dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) | |
242 | { | |
243 | dnode_t *dn; | |
244 | dmu_buf_impl_t *db; | |
245 | int error; | |
246 | ||
428870ff | 247 | error = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
248 | if (error) |
249 | return (error); | |
250 | ||
251 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
252 | if (dn->dn_bonus == NULL) { | |
253 | rw_exit(&dn->dn_struct_rwlock); | |
254 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
255 | if (dn->dn_bonus == NULL) | |
256 | dbuf_create_bonus(dn); | |
257 | } | |
258 | db = dn->dn_bonus; | |
34dc7c2f BB |
259 | |
260 | /* as long as the bonus buf is held, the dnode will be held */ | |
572e2857 | 261 | if (refcount_add(&db->db_holds, tag) == 1) { |
34dc7c2f | 262 | VERIFY(dnode_add_ref(dn, db)); |
572e2857 BB |
263 | (void) atomic_inc_32_nv(&dn->dn_dbufs_count); |
264 | } | |
265 | ||
266 | /* | |
267 | * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's | |
268 | * hold and incrementing the dbuf count to ensure that dnode_move() sees | |
269 | * a dnode hold for every dbuf. | |
270 | */ | |
271 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
272 | |
273 | dnode_rele(dn, FTAG); | |
274 | ||
428870ff | 275 | VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); |
34dc7c2f BB |
276 | |
277 | *dbp = &db->db; | |
278 | return (0); | |
279 | } | |
280 | ||
428870ff BB |
281 | /* |
282 | * returns ENOENT, EIO, or 0. | |
283 | * | |
284 | * This interface will allocate a blank spill dbuf when a spill blk | |
285 | * doesn't already exist on the dnode. | |
286 | * | |
287 | * if you only want to find an already existing spill db, then | |
288 | * dmu_spill_hold_existing() should be used. | |
289 | */ | |
290 | int | |
291 | dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) | |
292 | { | |
293 | dmu_buf_impl_t *db = NULL; | |
294 | int err; | |
295 | ||
296 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
297 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
298 | ||
299 | db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); | |
300 | ||
301 | if ((flags & DB_RF_HAVESTRUCT) == 0) | |
302 | rw_exit(&dn->dn_struct_rwlock); | |
303 | ||
304 | ASSERT(db != NULL); | |
572e2857 BB |
305 | err = dbuf_read(db, NULL, flags); |
306 | if (err == 0) | |
307 | *dbp = &db->db; | |
308 | else | |
309 | dbuf_rele(db, tag); | |
428870ff BB |
310 | return (err); |
311 | } | |
312 | ||
313 | int | |
314 | dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
315 | { | |
572e2857 BB |
316 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
317 | dnode_t *dn; | |
428870ff BB |
318 | int err; |
319 | ||
572e2857 BB |
320 | DB_DNODE_ENTER(db); |
321 | dn = DB_DNODE(db); | |
322 | ||
323 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { | |
324 | err = EINVAL; | |
325 | } else { | |
326 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
327 | ||
328 | if (!dn->dn_have_spill) { | |
329 | err = ENOENT; | |
330 | } else { | |
331 | err = dmu_spill_hold_by_dnode(dn, | |
332 | DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); | |
333 | } | |
428870ff | 334 | |
428870ff | 335 | rw_exit(&dn->dn_struct_rwlock); |
428870ff | 336 | } |
572e2857 BB |
337 | |
338 | DB_DNODE_EXIT(db); | |
428870ff BB |
339 | return (err); |
340 | } | |
341 | ||
342 | int | |
343 | dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) | |
344 | { | |
572e2857 BB |
345 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; |
346 | dnode_t *dn; | |
347 | int err; | |
348 | ||
349 | DB_DNODE_ENTER(db); | |
350 | dn = DB_DNODE(db); | |
351 | err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); | |
352 | DB_DNODE_EXIT(db); | |
353 | ||
354 | return (err); | |
428870ff BB |
355 | } |
356 | ||
34dc7c2f BB |
357 | /* |
358 | * Note: longer-term, we should modify all of the dmu_buf_*() interfaces | |
359 | * to take a held dnode rather than <os, object> -- the lookup is wasteful, | |
360 | * and can induce severe lock contention when writing to several files | |
361 | * whose dnodes are in the same block. | |
362 | */ | |
363 | static int | |
9babb374 BB |
364 | dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, |
365 | int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) | |
34dc7c2f | 366 | { |
b128c09f | 367 | dsl_pool_t *dp = NULL; |
34dc7c2f BB |
368 | dmu_buf_t **dbp; |
369 | uint64_t blkid, nblks, i; | |
9babb374 | 370 | uint32_t dbuf_flags; |
34dc7c2f BB |
371 | int err; |
372 | zio_t *zio; | |
d4ed6673 | 373 | hrtime_t start = 0; |
34dc7c2f BB |
374 | |
375 | ASSERT(length <= DMU_MAX_ACCESS); | |
376 | ||
45d1cae3 | 377 | dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT; |
9babb374 BB |
378 | if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz) |
379 | dbuf_flags |= DB_RF_NOPREFETCH; | |
34dc7c2f BB |
380 | |
381 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
382 | if (dn->dn_datablkshift) { | |
383 | int blkshift = dn->dn_datablkshift; | |
384 | nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - | |
385 | P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; | |
386 | } else { | |
387 | if (offset + length > dn->dn_datablksz) { | |
388 | zfs_panic_recover("zfs: accessing past end of object " | |
389 | "%llx/%llx (size=%u access=%llu+%llu)", | |
390 | (longlong_t)dn->dn_objset-> | |
391 | os_dsl_dataset->ds_object, | |
392 | (longlong_t)dn->dn_object, dn->dn_datablksz, | |
393 | (longlong_t)offset, (longlong_t)length); | |
45d1cae3 | 394 | rw_exit(&dn->dn_struct_rwlock); |
34dc7c2f BB |
395 | return (EIO); |
396 | } | |
397 | nblks = 1; | |
398 | } | |
b8d06fca | 399 | dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_PUSHPAGE | KM_NODEBUG); |
34dc7c2f | 400 | |
b128c09f BB |
401 | if (dn->dn_objset->os_dsl_dataset) |
402 | dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool; | |
403 | if (dp && dsl_pool_sync_context(dp)) | |
404 | start = gethrtime(); | |
405 | zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); | |
34dc7c2f BB |
406 | blkid = dbuf_whichblock(dn, offset); |
407 | for (i = 0; i < nblks; i++) { | |
408 | dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); | |
409 | if (db == NULL) { | |
410 | rw_exit(&dn->dn_struct_rwlock); | |
411 | dmu_buf_rele_array(dbp, nblks, tag); | |
412 | zio_nowait(zio); | |
413 | return (EIO); | |
414 | } | |
415 | /* initiate async i/o */ | |
416 | if (read) { | |
9babb374 | 417 | (void) dbuf_read(db, zio, dbuf_flags); |
34dc7c2f BB |
418 | } |
419 | dbp[i] = &db->db; | |
420 | } | |
421 | rw_exit(&dn->dn_struct_rwlock); | |
422 | ||
423 | /* wait for async i/o */ | |
424 | err = zio_wait(zio); | |
b128c09f BB |
425 | /* track read overhead when we are in sync context */ |
426 | if (dp && dsl_pool_sync_context(dp)) | |
427 | dp->dp_read_overhead += gethrtime() - start; | |
34dc7c2f BB |
428 | if (err) { |
429 | dmu_buf_rele_array(dbp, nblks, tag); | |
430 | return (err); | |
431 | } | |
432 | ||
433 | /* wait for other io to complete */ | |
434 | if (read) { | |
435 | for (i = 0; i < nblks; i++) { | |
436 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; | |
437 | mutex_enter(&db->db_mtx); | |
438 | while (db->db_state == DB_READ || | |
439 | db->db_state == DB_FILL) | |
440 | cv_wait(&db->db_changed, &db->db_mtx); | |
441 | if (db->db_state == DB_UNCACHED) | |
442 | err = EIO; | |
443 | mutex_exit(&db->db_mtx); | |
444 | if (err) { | |
445 | dmu_buf_rele_array(dbp, nblks, tag); | |
446 | return (err); | |
447 | } | |
448 | } | |
449 | } | |
450 | ||
451 | *numbufsp = nblks; | |
452 | *dbpp = dbp; | |
453 | return (0); | |
454 | } | |
455 | ||
456 | static int | |
457 | dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, | |
458 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) | |
459 | { | |
460 | dnode_t *dn; | |
461 | int err; | |
462 | ||
428870ff | 463 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
464 | if (err) |
465 | return (err); | |
466 | ||
467 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, | |
9babb374 | 468 | numbufsp, dbpp, DMU_READ_PREFETCH); |
34dc7c2f BB |
469 | |
470 | dnode_rele(dn, FTAG); | |
471 | ||
472 | return (err); | |
473 | } | |
474 | ||
475 | int | |
572e2857 | 476 | dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, |
34dc7c2f BB |
477 | uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) |
478 | { | |
572e2857 BB |
479 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
480 | dnode_t *dn; | |
34dc7c2f BB |
481 | int err; |
482 | ||
572e2857 BB |
483 | DB_DNODE_ENTER(db); |
484 | dn = DB_DNODE(db); | |
34dc7c2f | 485 | err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, |
9babb374 | 486 | numbufsp, dbpp, DMU_READ_PREFETCH); |
572e2857 | 487 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
488 | |
489 | return (err); | |
490 | } | |
491 | ||
492 | void | |
493 | dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) | |
494 | { | |
495 | int i; | |
496 | dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; | |
497 | ||
498 | if (numbufs == 0) | |
499 | return; | |
500 | ||
501 | for (i = 0; i < numbufs; i++) { | |
502 | if (dbp[i]) | |
503 | dbuf_rele(dbp[i], tag); | |
504 | } | |
505 | ||
506 | kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); | |
507 | } | |
508 | ||
509 | void | |
510 | dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) | |
511 | { | |
512 | dnode_t *dn; | |
513 | uint64_t blkid; | |
514 | int nblks, i, err; | |
515 | ||
516 | if (zfs_prefetch_disable) | |
517 | return; | |
518 | ||
519 | if (len == 0) { /* they're interested in the bonus buffer */ | |
572e2857 | 520 | dn = DMU_META_DNODE(os); |
34dc7c2f BB |
521 | |
522 | if (object == 0 || object >= DN_MAX_OBJECT) | |
523 | return; | |
524 | ||
525 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
526 | blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); | |
527 | dbuf_prefetch(dn, blkid); | |
528 | rw_exit(&dn->dn_struct_rwlock); | |
529 | return; | |
530 | } | |
531 | ||
532 | /* | |
533 | * XXX - Note, if the dnode for the requested object is not | |
534 | * already cached, we will do a *synchronous* read in the | |
535 | * dnode_hold() call. The same is true for any indirects. | |
536 | */ | |
428870ff | 537 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
538 | if (err != 0) |
539 | return; | |
540 | ||
541 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
542 | if (dn->dn_datablkshift) { | |
543 | int blkshift = dn->dn_datablkshift; | |
544 | nblks = (P2ROUNDUP(offset+len, 1<<blkshift) - | |
545 | P2ALIGN(offset, 1<<blkshift)) >> blkshift; | |
546 | } else { | |
547 | nblks = (offset < dn->dn_datablksz); | |
548 | } | |
549 | ||
550 | if (nblks != 0) { | |
551 | blkid = dbuf_whichblock(dn, offset); | |
552 | for (i = 0; i < nblks; i++) | |
553 | dbuf_prefetch(dn, blkid+i); | |
554 | } | |
555 | ||
556 | rw_exit(&dn->dn_struct_rwlock); | |
557 | ||
558 | dnode_rele(dn, FTAG); | |
559 | } | |
560 | ||
45d1cae3 BB |
561 | /* |
562 | * Get the next "chunk" of file data to free. We traverse the file from | |
563 | * the end so that the file gets shorter over time (if we crashes in the | |
564 | * middle, this will leave us in a better state). We find allocated file | |
565 | * data by simply searching the allocated level 1 indirects. | |
566 | */ | |
b128c09f | 567 | static int |
45d1cae3 | 568 | get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit) |
b128c09f | 569 | { |
45d1cae3 BB |
570 | uint64_t len = *start - limit; |
571 | uint64_t blkcnt = 0; | |
572 | uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1)); | |
573 | uint64_t iblkrange = | |
b128c09f BB |
574 | dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); |
575 | ||
45d1cae3 | 576 | ASSERT(limit <= *start); |
b128c09f | 577 | |
45d1cae3 BB |
578 | if (len <= iblkrange * maxblks) { |
579 | *start = limit; | |
b128c09f BB |
580 | return (0); |
581 | } | |
45d1cae3 | 582 | ASSERT(ISP2(iblkrange)); |
b128c09f | 583 | |
45d1cae3 | 584 | while (*start > limit && blkcnt < maxblks) { |
b128c09f BB |
585 | int err; |
586 | ||
45d1cae3 | 587 | /* find next allocated L1 indirect */ |
b128c09f | 588 | err = dnode_next_offset(dn, |
45d1cae3 | 589 | DNODE_FIND_BACKWARDS, start, 2, 1, 0); |
b128c09f | 590 | |
45d1cae3 BB |
591 | /* if there are no more, then we are done */ |
592 | if (err == ESRCH) { | |
593 | *start = limit; | |
b128c09f | 594 | return (0); |
45d1cae3 | 595 | } else if (err) { |
b128c09f | 596 | return (err); |
45d1cae3 BB |
597 | } |
598 | blkcnt += 1; | |
b128c09f | 599 | |
45d1cae3 BB |
600 | /* reset offset to end of "next" block back */ |
601 | *start = P2ALIGN(*start, iblkrange); | |
602 | if (*start <= limit) | |
603 | *start = limit; | |
604 | else | |
605 | *start -= 1; | |
b128c09f BB |
606 | } |
607 | return (0); | |
608 | } | |
609 | ||
610 | static int | |
611 | dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, | |
612 | uint64_t length, boolean_t free_dnode) | |
613 | { | |
614 | dmu_tx_t *tx; | |
615 | uint64_t object_size, start, end, len; | |
616 | boolean_t trunc = (length == DMU_OBJECT_END); | |
617 | int align, err; | |
618 | ||
619 | align = 1 << dn->dn_datablkshift; | |
620 | ASSERT(align > 0); | |
621 | object_size = align == 1 ? dn->dn_datablksz : | |
622 | (dn->dn_maxblkid + 1) << dn->dn_datablkshift; | |
623 | ||
9babb374 BB |
624 | end = offset + length; |
625 | if (trunc || end > object_size) | |
b128c09f BB |
626 | end = object_size; |
627 | if (end <= offset) | |
628 | return (0); | |
629 | length = end - offset; | |
630 | ||
631 | while (length) { | |
632 | start = end; | |
9babb374 | 633 | /* assert(offset <= start) */ |
b128c09f BB |
634 | err = get_next_chunk(dn, &start, offset); |
635 | if (err) | |
636 | return (err); | |
637 | len = trunc ? DMU_OBJECT_END : end - start; | |
638 | ||
639 | tx = dmu_tx_create(os); | |
640 | dmu_tx_hold_free(tx, dn->dn_object, start, len); | |
641 | err = dmu_tx_assign(tx, TXG_WAIT); | |
642 | if (err) { | |
643 | dmu_tx_abort(tx); | |
644 | return (err); | |
645 | } | |
646 | ||
647 | dnode_free_range(dn, start, trunc ? -1 : len, tx); | |
648 | ||
649 | if (start == 0 && free_dnode) { | |
650 | ASSERT(trunc); | |
651 | dnode_free(dn, tx); | |
652 | } | |
653 | ||
654 | length -= end - start; | |
655 | ||
656 | dmu_tx_commit(tx); | |
657 | end = start; | |
658 | } | |
659 | return (0); | |
660 | } | |
661 | ||
662 | int | |
663 | dmu_free_long_range(objset_t *os, uint64_t object, | |
664 | uint64_t offset, uint64_t length) | |
665 | { | |
666 | dnode_t *dn; | |
667 | int err; | |
668 | ||
428870ff | 669 | err = dnode_hold(os, object, FTAG, &dn); |
b128c09f BB |
670 | if (err != 0) |
671 | return (err); | |
672 | err = dmu_free_long_range_impl(os, dn, offset, length, FALSE); | |
673 | dnode_rele(dn, FTAG); | |
674 | return (err); | |
675 | } | |
676 | ||
677 | int | |
678 | dmu_free_object(objset_t *os, uint64_t object) | |
679 | { | |
680 | dnode_t *dn; | |
681 | dmu_tx_t *tx; | |
682 | int err; | |
683 | ||
428870ff | 684 | err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, |
b128c09f BB |
685 | FTAG, &dn); |
686 | if (err != 0) | |
687 | return (err); | |
688 | if (dn->dn_nlevels == 1) { | |
689 | tx = dmu_tx_create(os); | |
690 | dmu_tx_hold_bonus(tx, object); | |
691 | dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END); | |
692 | err = dmu_tx_assign(tx, TXG_WAIT); | |
693 | if (err == 0) { | |
694 | dnode_free_range(dn, 0, DMU_OBJECT_END, tx); | |
695 | dnode_free(dn, tx); | |
696 | dmu_tx_commit(tx); | |
697 | } else { | |
698 | dmu_tx_abort(tx); | |
699 | } | |
700 | } else { | |
701 | err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE); | |
702 | } | |
703 | dnode_rele(dn, FTAG); | |
704 | return (err); | |
705 | } | |
706 | ||
34dc7c2f BB |
707 | int |
708 | dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, | |
709 | uint64_t size, dmu_tx_t *tx) | |
710 | { | |
711 | dnode_t *dn; | |
428870ff | 712 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
713 | if (err) |
714 | return (err); | |
715 | ASSERT(offset < UINT64_MAX); | |
716 | ASSERT(size == -1ULL || size <= UINT64_MAX - offset); | |
717 | dnode_free_range(dn, offset, size, tx); | |
718 | dnode_rele(dn, FTAG); | |
719 | return (0); | |
720 | } | |
721 | ||
722 | int | |
723 | dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
9babb374 | 724 | void *buf, uint32_t flags) |
34dc7c2f BB |
725 | { |
726 | dnode_t *dn; | |
727 | dmu_buf_t **dbp; | |
45d1cae3 | 728 | int numbufs, err; |
34dc7c2f | 729 | |
428870ff | 730 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
731 | if (err) |
732 | return (err); | |
733 | ||
734 | /* | |
735 | * Deal with odd block sizes, where there can't be data past the first | |
736 | * block. If we ever do the tail block optimization, we will need to | |
737 | * handle that here as well. | |
738 | */ | |
45d1cae3 | 739 | if (dn->dn_maxblkid == 0) { |
34dc7c2f BB |
740 | int newsz = offset > dn->dn_datablksz ? 0 : |
741 | MIN(size, dn->dn_datablksz - offset); | |
742 | bzero((char *)buf + newsz, size - newsz); | |
743 | size = newsz; | |
744 | } | |
745 | ||
746 | while (size > 0) { | |
747 | uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); | |
45d1cae3 | 748 | int i; |
34dc7c2f BB |
749 | |
750 | /* | |
751 | * NB: we could do this block-at-a-time, but it's nice | |
752 | * to be reading in parallel. | |
753 | */ | |
754 | err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, | |
9babb374 | 755 | TRUE, FTAG, &numbufs, &dbp, flags); |
34dc7c2f BB |
756 | if (err) |
757 | break; | |
758 | ||
759 | for (i = 0; i < numbufs; i++) { | |
760 | int tocpy; | |
761 | int bufoff; | |
762 | dmu_buf_t *db = dbp[i]; | |
763 | ||
764 | ASSERT(size > 0); | |
765 | ||
766 | bufoff = offset - db->db_offset; | |
767 | tocpy = (int)MIN(db->db_size - bufoff, size); | |
768 | ||
769 | bcopy((char *)db->db_data + bufoff, buf, tocpy); | |
770 | ||
771 | offset += tocpy; | |
772 | size -= tocpy; | |
773 | buf = (char *)buf + tocpy; | |
774 | } | |
775 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
776 | } | |
777 | dnode_rele(dn, FTAG); | |
778 | return (err); | |
779 | } | |
780 | ||
781 | void | |
782 | dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
783 | const void *buf, dmu_tx_t *tx) | |
784 | { | |
785 | dmu_buf_t **dbp; | |
786 | int numbufs, i; | |
787 | ||
788 | if (size == 0) | |
789 | return; | |
790 | ||
791 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
792 | FALSE, FTAG, &numbufs, &dbp)); | |
793 | ||
794 | for (i = 0; i < numbufs; i++) { | |
795 | int tocpy; | |
796 | int bufoff; | |
797 | dmu_buf_t *db = dbp[i]; | |
798 | ||
799 | ASSERT(size > 0); | |
800 | ||
801 | bufoff = offset - db->db_offset; | |
802 | tocpy = (int)MIN(db->db_size - bufoff, size); | |
803 | ||
804 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
805 | ||
806 | if (tocpy == db->db_size) | |
807 | dmu_buf_will_fill(db, tx); | |
808 | else | |
809 | dmu_buf_will_dirty(db, tx); | |
810 | ||
60101509 | 811 | (void) memcpy((char *)db->db_data + bufoff, buf, tocpy); |
34dc7c2f BB |
812 | |
813 | if (tocpy == db->db_size) | |
814 | dmu_buf_fill_done(db, tx); | |
815 | ||
816 | offset += tocpy; | |
817 | size -= tocpy; | |
818 | buf = (char *)buf + tocpy; | |
819 | } | |
820 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
821 | } | |
822 | ||
b128c09f BB |
823 | void |
824 | dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, | |
825 | dmu_tx_t *tx) | |
826 | { | |
827 | dmu_buf_t **dbp; | |
828 | int numbufs, i; | |
829 | ||
830 | if (size == 0) | |
831 | return; | |
832 | ||
833 | VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, | |
834 | FALSE, FTAG, &numbufs, &dbp)); | |
835 | ||
836 | for (i = 0; i < numbufs; i++) { | |
837 | dmu_buf_t *db = dbp[i]; | |
838 | ||
839 | dmu_buf_will_not_fill(db, tx); | |
840 | } | |
841 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
842 | } | |
843 | ||
428870ff BB |
844 | /* |
845 | * DMU support for xuio | |
846 | */ | |
847 | kstat_t *xuio_ksp = NULL; | |
848 | ||
59e6e7ca BB |
849 | typedef struct xuio_stats { |
850 | /* loaned yet not returned arc_buf */ | |
851 | kstat_named_t xuiostat_onloan_rbuf; | |
852 | kstat_named_t xuiostat_onloan_wbuf; | |
853 | /* whether a copy is made when loaning out a read buffer */ | |
854 | kstat_named_t xuiostat_rbuf_copied; | |
855 | kstat_named_t xuiostat_rbuf_nocopy; | |
856 | /* whether a copy is made when assigning a write buffer */ | |
857 | kstat_named_t xuiostat_wbuf_copied; | |
858 | kstat_named_t xuiostat_wbuf_nocopy; | |
859 | } xuio_stats_t; | |
860 | ||
861 | static xuio_stats_t xuio_stats = { | |
862 | { "onloan_read_buf", KSTAT_DATA_UINT64 }, | |
863 | { "onloan_write_buf", KSTAT_DATA_UINT64 }, | |
864 | { "read_buf_copied", KSTAT_DATA_UINT64 }, | |
865 | { "read_buf_nocopy", KSTAT_DATA_UINT64 }, | |
866 | { "write_buf_copied", KSTAT_DATA_UINT64 }, | |
867 | { "write_buf_nocopy", KSTAT_DATA_UINT64 } | |
868 | }; | |
869 | ||
870 | #define XUIOSTAT_INCR(stat, val) \ | |
871 | atomic_add_64(&xuio_stats.stat.value.ui64, (val)) | |
872 | #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1) | |
873 | ||
428870ff BB |
874 | int |
875 | dmu_xuio_init(xuio_t *xuio, int nblk) | |
876 | { | |
877 | dmu_xuio_t *priv; | |
878 | uio_t *uio = &xuio->xu_uio; | |
879 | ||
880 | uio->uio_iovcnt = nblk; | |
b8d06fca | 881 | uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_PUSHPAGE); |
428870ff | 882 | |
b8d06fca | 883 | priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_PUSHPAGE); |
428870ff | 884 | priv->cnt = nblk; |
b8d06fca | 885 | priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_PUSHPAGE); |
428870ff BB |
886 | priv->iovp = uio->uio_iov; |
887 | XUIO_XUZC_PRIV(xuio) = priv; | |
888 | ||
889 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
890 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); | |
891 | else | |
892 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); | |
893 | ||
894 | return (0); | |
895 | } | |
896 | ||
897 | void | |
898 | dmu_xuio_fini(xuio_t *xuio) | |
899 | { | |
900 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
901 | int nblk = priv->cnt; | |
902 | ||
903 | kmem_free(priv->iovp, nblk * sizeof (iovec_t)); | |
904 | kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); | |
905 | kmem_free(priv, sizeof (dmu_xuio_t)); | |
906 | ||
907 | if (XUIO_XUZC_RW(xuio) == UIO_READ) | |
908 | XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); | |
909 | else | |
910 | XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); | |
911 | } | |
912 | ||
913 | /* | |
914 | * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } | |
915 | * and increase priv->next by 1. | |
916 | */ | |
917 | int | |
918 | dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) | |
919 | { | |
920 | struct iovec *iov; | |
921 | uio_t *uio = &xuio->xu_uio; | |
922 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
923 | int i = priv->next++; | |
924 | ||
925 | ASSERT(i < priv->cnt); | |
926 | ASSERT(off + n <= arc_buf_size(abuf)); | |
927 | iov = uio->uio_iov + i; | |
928 | iov->iov_base = (char *)abuf->b_data + off; | |
929 | iov->iov_len = n; | |
930 | priv->bufs[i] = abuf; | |
931 | return (0); | |
932 | } | |
933 | ||
934 | int | |
935 | dmu_xuio_cnt(xuio_t *xuio) | |
936 | { | |
937 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
938 | return (priv->cnt); | |
939 | } | |
940 | ||
941 | arc_buf_t * | |
942 | dmu_xuio_arcbuf(xuio_t *xuio, int i) | |
943 | { | |
944 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
945 | ||
946 | ASSERT(i < priv->cnt); | |
947 | return (priv->bufs[i]); | |
948 | } | |
949 | ||
950 | void | |
951 | dmu_xuio_clear(xuio_t *xuio, int i) | |
952 | { | |
953 | dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); | |
954 | ||
955 | ASSERT(i < priv->cnt); | |
956 | priv->bufs[i] = NULL; | |
957 | } | |
958 | ||
959 | static void | |
960 | xuio_stat_init(void) | |
961 | { | |
962 | xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", | |
963 | KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), | |
964 | KSTAT_FLAG_VIRTUAL); | |
965 | if (xuio_ksp != NULL) { | |
966 | xuio_ksp->ks_data = &xuio_stats; | |
967 | kstat_install(xuio_ksp); | |
968 | } | |
969 | } | |
970 | ||
971 | static void | |
972 | xuio_stat_fini(void) | |
973 | { | |
974 | if (xuio_ksp != NULL) { | |
975 | kstat_delete(xuio_ksp); | |
976 | xuio_ksp = NULL; | |
977 | } | |
978 | } | |
979 | ||
980 | void | |
981 | xuio_stat_wbuf_copied() | |
982 | { | |
983 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); | |
984 | } | |
985 | ||
986 | void | |
987 | xuio_stat_wbuf_nocopy() | |
988 | { | |
989 | XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); | |
990 | } | |
991 | ||
34dc7c2f | 992 | #ifdef _KERNEL |
60101509 BB |
993 | |
994 | /* | |
995 | * Copy up to size bytes between arg_buf and req based on the data direction | |
996 | * described by the req. If an entire req's data cannot be transfered the | |
997 | * req's is updated such that it's current index and bv offsets correctly | |
998 | * reference any residual data which could not be copied. The return value | |
999 | * is the number of bytes successfully copied to arg_buf. | |
1000 | */ | |
1001 | static int | |
1002 | dmu_req_copy(void *arg_buf, int size, int *offset, struct request *req) | |
1003 | { | |
1004 | struct bio_vec *bv; | |
1005 | struct req_iterator iter; | |
1006 | char *bv_buf; | |
1007 | int tocpy; | |
1008 | ||
1009 | *offset = 0; | |
1010 | rq_for_each_segment(bv, req, iter) { | |
1011 | ||
1012 | /* Fully consumed the passed arg_buf */ | |
1013 | ASSERT3S(*offset, <=, size); | |
1014 | if (size == *offset) | |
1015 | break; | |
1016 | ||
1017 | /* Skip fully consumed bv's */ | |
1018 | if (bv->bv_len == 0) | |
1019 | continue; | |
1020 | ||
1021 | tocpy = MIN(bv->bv_len, size - *offset); | |
1022 | ASSERT3S(tocpy, >=, 0); | |
1023 | ||
1024 | bv_buf = page_address(bv->bv_page) + bv->bv_offset; | |
1025 | ASSERT3P(bv_buf, !=, NULL); | |
1026 | ||
1027 | if (rq_data_dir(req) == WRITE) | |
1028 | memcpy(arg_buf + *offset, bv_buf, tocpy); | |
1029 | else | |
1030 | memcpy(bv_buf, arg_buf + *offset, tocpy); | |
1031 | ||
1032 | *offset += tocpy; | |
1033 | bv->bv_offset += tocpy; | |
1034 | bv->bv_len -= tocpy; | |
1035 | } | |
1036 | ||
1037 | return 0; | |
1038 | } | |
1039 | ||
e26ade51 BB |
1040 | static void |
1041 | dmu_bio_put(struct bio *bio) | |
1042 | { | |
1043 | struct bio *bio_next; | |
1044 | ||
1045 | while (bio) { | |
1046 | bio_next = bio->bi_next; | |
1047 | bio_put(bio); | |
1048 | bio = bio_next; | |
1049 | } | |
1050 | } | |
1051 | ||
1052 | static int | |
1053 | dmu_bio_clone(struct bio *bio, struct bio **bio_copy) | |
1054 | { | |
1055 | struct bio *bio_root = NULL; | |
1056 | struct bio *bio_last = NULL; | |
1057 | struct bio *bio_new; | |
1058 | ||
1059 | if (bio == NULL) | |
1060 | return EINVAL; | |
1061 | ||
1062 | while (bio) { | |
1063 | bio_new = bio_clone(bio, GFP_NOIO); | |
1064 | if (bio_new == NULL) { | |
1065 | dmu_bio_put(bio_root); | |
1066 | return ENOMEM; | |
1067 | } | |
1068 | ||
1069 | if (bio_last) { | |
1070 | bio_last->bi_next = bio_new; | |
1071 | bio_last = bio_new; | |
1072 | } else { | |
1073 | bio_root = bio_new; | |
1074 | bio_last = bio_new; | |
1075 | } | |
1076 | ||
1077 | bio = bio->bi_next; | |
1078 | } | |
1079 | ||
1080 | *bio_copy = bio_root; | |
1081 | ||
1082 | return 0; | |
1083 | } | |
1084 | ||
34dc7c2f | 1085 | int |
60101509 | 1086 | dmu_read_req(objset_t *os, uint64_t object, struct request *req) |
34dc7c2f | 1087 | { |
60101509 BB |
1088 | uint64_t size = blk_rq_bytes(req); |
1089 | uint64_t offset = blk_rq_pos(req) << 9; | |
e26ade51 | 1090 | struct bio *bio_saved = req->bio; |
34dc7c2f BB |
1091 | dmu_buf_t **dbp; |
1092 | int numbufs, i, err; | |
1093 | ||
1094 | /* | |
1095 | * NB: we could do this block-at-a-time, but it's nice | |
1096 | * to be reading in parallel. | |
1097 | */ | |
60101509 BB |
1098 | err = dmu_buf_hold_array(os, object, offset, size, TRUE, FTAG, |
1099 | &numbufs, &dbp); | |
34dc7c2f BB |
1100 | if (err) |
1101 | return (err); | |
1102 | ||
e26ade51 BB |
1103 | /* |
1104 | * Clone the bio list so the bv->bv_offset and bv->bv_len members | |
1105 | * can be safely modified. The original bio list is relinked in to | |
1106 | * the request when the function exits. This is required because | |
1107 | * some file systems blindly assume that these values will remain | |
1108 | * constant between bio_submit() and the IO completion callback. | |
1109 | */ | |
1110 | err = dmu_bio_clone(bio_saved, &req->bio); | |
1111 | if (err) | |
1112 | goto error; | |
1113 | ||
34dc7c2f | 1114 | for (i = 0; i < numbufs; i++) { |
60101509 | 1115 | int tocpy, didcpy, bufoff; |
34dc7c2f BB |
1116 | dmu_buf_t *db = dbp[i]; |
1117 | ||
60101509 BB |
1118 | bufoff = offset - db->db_offset; |
1119 | ASSERT3S(bufoff, >=, 0); | |
34dc7c2f | 1120 | |
34dc7c2f | 1121 | tocpy = (int)MIN(db->db_size - bufoff, size); |
60101509 BB |
1122 | if (tocpy == 0) |
1123 | break; | |
34dc7c2f | 1124 | |
60101509 BB |
1125 | err = dmu_req_copy(db->db_data + bufoff, tocpy, &didcpy, req); |
1126 | ||
1127 | if (didcpy < tocpy) | |
1128 | err = EIO; | |
428870ff | 1129 | |
34dc7c2f BB |
1130 | if (err) |
1131 | break; | |
1132 | ||
1133 | size -= tocpy; | |
60101509 BB |
1134 | offset += didcpy; |
1135 | err = 0; | |
34dc7c2f | 1136 | } |
e26ade51 BB |
1137 | |
1138 | dmu_bio_put(req->bio); | |
1139 | req->bio = bio_saved; | |
1140 | error: | |
34dc7c2f BB |
1141 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
1142 | ||
1143 | return (err); | |
1144 | } | |
1145 | ||
60101509 BB |
1146 | int |
1147 | dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx) | |
34dc7c2f | 1148 | { |
60101509 BB |
1149 | uint64_t size = blk_rq_bytes(req); |
1150 | uint64_t offset = blk_rq_pos(req) << 9; | |
e26ade51 | 1151 | struct bio *bio_saved = req->bio; |
34dc7c2f | 1152 | dmu_buf_t **dbp; |
428870ff | 1153 | int numbufs; |
34dc7c2f | 1154 | int err = 0; |
428870ff | 1155 | int i; |
34dc7c2f | 1156 | |
60101509 BB |
1157 | if (size == 0) |
1158 | return (0); | |
1159 | ||
1160 | err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, | |
1161 | &numbufs, &dbp); | |
34dc7c2f BB |
1162 | if (err) |
1163 | return (err); | |
1164 | ||
e26ade51 BB |
1165 | /* |
1166 | * Clone the bio list so the bv->bv_offset and bv->bv_len members | |
1167 | * can be safely modified. The original bio list is relinked in to | |
1168 | * the request when the function exits. This is required because | |
1169 | * some file systems blindly assume that these values will remain | |
1170 | * constant between bio_submit() and the IO completion callback. | |
1171 | */ | |
1172 | err = dmu_bio_clone(bio_saved, &req->bio); | |
1173 | if (err) | |
1174 | goto error; | |
1175 | ||
34dc7c2f | 1176 | for (i = 0; i < numbufs; i++) { |
60101509 | 1177 | int tocpy, didcpy, bufoff; |
34dc7c2f BB |
1178 | dmu_buf_t *db = dbp[i]; |
1179 | ||
60101509 BB |
1180 | bufoff = offset - db->db_offset; |
1181 | ASSERT3S(bufoff, >=, 0); | |
34dc7c2f | 1182 | |
34dc7c2f | 1183 | tocpy = (int)MIN(db->db_size - bufoff, size); |
60101509 BB |
1184 | if (tocpy == 0) |
1185 | break; | |
34dc7c2f BB |
1186 | |
1187 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1188 | ||
1189 | if (tocpy == db->db_size) | |
1190 | dmu_buf_will_fill(db, tx); | |
1191 | else | |
1192 | dmu_buf_will_dirty(db, tx); | |
1193 | ||
60101509 | 1194 | err = dmu_req_copy(db->db_data + bufoff, tocpy, &didcpy, req); |
34dc7c2f BB |
1195 | |
1196 | if (tocpy == db->db_size) | |
1197 | dmu_buf_fill_done(db, tx); | |
1198 | ||
60101509 BB |
1199 | if (didcpy < tocpy) |
1200 | err = EIO; | |
1201 | ||
34dc7c2f BB |
1202 | if (err) |
1203 | break; | |
1204 | ||
1205 | size -= tocpy; | |
60101509 BB |
1206 | offset += didcpy; |
1207 | err = 0; | |
34dc7c2f | 1208 | } |
428870ff | 1209 | |
e26ade51 BB |
1210 | dmu_bio_put(req->bio); |
1211 | req->bio = bio_saved; | |
1212 | error: | |
34dc7c2f | 1213 | dmu_buf_rele_array(dbp, numbufs, FTAG); |
e26ade51 | 1214 | |
34dc7c2f BB |
1215 | return (err); |
1216 | } | |
1217 | ||
872e8d26 BB |
1218 | int |
1219 | dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) | |
1220 | { | |
1221 | dmu_buf_t **dbp; | |
1222 | int numbufs, i, err; | |
1223 | xuio_t *xuio = NULL; | |
1224 | ||
1225 | /* | |
1226 | * NB: we could do this block-at-a-time, but it's nice | |
1227 | * to be reading in parallel. | |
1228 | */ | |
1229 | err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG, | |
1230 | &numbufs, &dbp); | |
1231 | if (err) | |
1232 | return (err); | |
1233 | ||
1234 | for (i = 0; i < numbufs; i++) { | |
1235 | int tocpy; | |
1236 | int bufoff; | |
1237 | dmu_buf_t *db = dbp[i]; | |
1238 | ||
1239 | ASSERT(size > 0); | |
1240 | ||
1241 | bufoff = uio->uio_loffset - db->db_offset; | |
1242 | tocpy = (int)MIN(db->db_size - bufoff, size); | |
1243 | ||
1244 | if (xuio) { | |
1245 | dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; | |
1246 | arc_buf_t *dbuf_abuf = dbi->db_buf; | |
1247 | arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); | |
1248 | err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); | |
1249 | if (!err) { | |
1250 | uio->uio_resid -= tocpy; | |
1251 | uio->uio_loffset += tocpy; | |
1252 | } | |
1253 | ||
1254 | if (abuf == dbuf_abuf) | |
1255 | XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); | |
1256 | else | |
1257 | XUIOSTAT_BUMP(xuiostat_rbuf_copied); | |
1258 | } else { | |
1259 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1260 | UIO_READ, uio); | |
1261 | } | |
1262 | if (err) | |
1263 | break; | |
1264 | ||
1265 | size -= tocpy; | |
1266 | } | |
1267 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1268 | ||
1269 | return (err); | |
1270 | } | |
1271 | ||
1272 | static int | |
1273 | dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) | |
1274 | { | |
1275 | dmu_buf_t **dbp; | |
1276 | int numbufs; | |
1277 | int err = 0; | |
1278 | int i; | |
1279 | ||
1280 | err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, | |
1281 | FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); | |
1282 | if (err) | |
1283 | return (err); | |
1284 | ||
1285 | for (i = 0; i < numbufs; i++) { | |
1286 | int tocpy; | |
1287 | int bufoff; | |
1288 | dmu_buf_t *db = dbp[i]; | |
1289 | ||
1290 | ASSERT(size > 0); | |
1291 | ||
1292 | bufoff = uio->uio_loffset - db->db_offset; | |
1293 | tocpy = (int)MIN(db->db_size - bufoff, size); | |
1294 | ||
1295 | ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); | |
1296 | ||
1297 | if (tocpy == db->db_size) | |
1298 | dmu_buf_will_fill(db, tx); | |
1299 | else | |
1300 | dmu_buf_will_dirty(db, tx); | |
1301 | ||
1302 | /* | |
1303 | * XXX uiomove could block forever (eg.nfs-backed | |
1304 | * pages). There needs to be a uiolockdown() function | |
1305 | * to lock the pages in memory, so that uiomove won't | |
1306 | * block. | |
1307 | */ | |
1308 | err = uiomove((char *)db->db_data + bufoff, tocpy, | |
1309 | UIO_WRITE, uio); | |
1310 | ||
1311 | if (tocpy == db->db_size) | |
1312 | dmu_buf_fill_done(db, tx); | |
1313 | ||
1314 | if (err) | |
1315 | break; | |
1316 | ||
1317 | size -= tocpy; | |
1318 | } | |
1319 | ||
1320 | dmu_buf_rele_array(dbp, numbufs, FTAG); | |
1321 | return (err); | |
1322 | } | |
1323 | ||
428870ff BB |
1324 | int |
1325 | dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, | |
1326 | dmu_tx_t *tx) | |
1327 | { | |
572e2857 BB |
1328 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; |
1329 | dnode_t *dn; | |
1330 | int err; | |
1331 | ||
428870ff BB |
1332 | if (size == 0) |
1333 | return (0); | |
1334 | ||
572e2857 BB |
1335 | DB_DNODE_ENTER(db); |
1336 | dn = DB_DNODE(db); | |
1337 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1338 | DB_DNODE_EXIT(db); | |
1339 | ||
1340 | return (err); | |
428870ff BB |
1341 | } |
1342 | ||
1343 | int | |
1344 | dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, | |
1345 | dmu_tx_t *tx) | |
1346 | { | |
1347 | dnode_t *dn; | |
1348 | int err; | |
1349 | ||
1350 | if (size == 0) | |
1351 | return (0); | |
1352 | ||
1353 | err = dnode_hold(os, object, FTAG, &dn); | |
1354 | if (err) | |
1355 | return (err); | |
1356 | ||
1357 | err = dmu_write_uio_dnode(dn, uio, size, tx); | |
1358 | ||
1359 | dnode_rele(dn, FTAG); | |
1360 | ||
1361 | return (err); | |
1362 | } | |
872e8d26 | 1363 | #endif /* _KERNEL */ |
34dc7c2f | 1364 | |
9babb374 BB |
1365 | /* |
1366 | * Allocate a loaned anonymous arc buffer. | |
1367 | */ | |
1368 | arc_buf_t * | |
1369 | dmu_request_arcbuf(dmu_buf_t *handle, int size) | |
1370 | { | |
572e2857 BB |
1371 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; |
1372 | spa_t *spa; | |
9babb374 | 1373 | |
572e2857 BB |
1374 | DB_GET_SPA(&spa, db); |
1375 | return (arc_loan_buf(spa, size)); | |
9babb374 BB |
1376 | } |
1377 | ||
1378 | /* | |
1379 | * Free a loaned arc buffer. | |
1380 | */ | |
1381 | void | |
1382 | dmu_return_arcbuf(arc_buf_t *buf) | |
1383 | { | |
1384 | arc_return_buf(buf, FTAG); | |
13fe0198 | 1385 | VERIFY(arc_buf_remove_ref(buf, FTAG)); |
9babb374 BB |
1386 | } |
1387 | ||
1388 | /* | |
1389 | * When possible directly assign passed loaned arc buffer to a dbuf. | |
1390 | * If this is not possible copy the contents of passed arc buf via | |
1391 | * dmu_write(). | |
1392 | */ | |
1393 | void | |
1394 | dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, | |
1395 | dmu_tx_t *tx) | |
1396 | { | |
572e2857 BB |
1397 | dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; |
1398 | dnode_t *dn; | |
9babb374 BB |
1399 | dmu_buf_impl_t *db; |
1400 | uint32_t blksz = (uint32_t)arc_buf_size(buf); | |
1401 | uint64_t blkid; | |
1402 | ||
572e2857 BB |
1403 | DB_DNODE_ENTER(dbuf); |
1404 | dn = DB_DNODE(dbuf); | |
9babb374 BB |
1405 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
1406 | blkid = dbuf_whichblock(dn, offset); | |
1407 | VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); | |
1408 | rw_exit(&dn->dn_struct_rwlock); | |
572e2857 | 1409 | DB_DNODE_EXIT(dbuf); |
9babb374 BB |
1410 | |
1411 | if (offset == db->db.db_offset && blksz == db->db.db_size) { | |
1412 | dbuf_assign_arcbuf(db, buf, tx); | |
1413 | dbuf_rele(db, FTAG); | |
1414 | } else { | |
572e2857 BB |
1415 | objset_t *os; |
1416 | uint64_t object; | |
1417 | ||
1418 | DB_DNODE_ENTER(dbuf); | |
1419 | dn = DB_DNODE(dbuf); | |
1420 | os = dn->dn_objset; | |
1421 | object = dn->dn_object; | |
1422 | DB_DNODE_EXIT(dbuf); | |
1423 | ||
9babb374 | 1424 | dbuf_rele(db, FTAG); |
572e2857 | 1425 | dmu_write(os, object, offset, blksz, buf->b_data, tx); |
9babb374 | 1426 | dmu_return_arcbuf(buf); |
428870ff | 1427 | XUIOSTAT_BUMP(xuiostat_wbuf_copied); |
9babb374 BB |
1428 | } |
1429 | } | |
1430 | ||
34dc7c2f | 1431 | typedef struct { |
428870ff BB |
1432 | dbuf_dirty_record_t *dsa_dr; |
1433 | dmu_sync_cb_t *dsa_done; | |
1434 | zgd_t *dsa_zgd; | |
1435 | dmu_tx_t *dsa_tx; | |
34dc7c2f BB |
1436 | } dmu_sync_arg_t; |
1437 | ||
b128c09f BB |
1438 | /* ARGSUSED */ |
1439 | static void | |
1440 | dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) | |
1441 | { | |
428870ff BB |
1442 | dmu_sync_arg_t *dsa = varg; |
1443 | dmu_buf_t *db = dsa->dsa_zgd->zgd_db; | |
b128c09f BB |
1444 | blkptr_t *bp = zio->io_bp; |
1445 | ||
428870ff BB |
1446 | if (zio->io_error == 0) { |
1447 | if (BP_IS_HOLE(bp)) { | |
1448 | /* | |
1449 | * A block of zeros may compress to a hole, but the | |
1450 | * block size still needs to be known for replay. | |
1451 | */ | |
1452 | BP_SET_LSIZE(bp, db->db_size); | |
1453 | } else { | |
428870ff BB |
1454 | ASSERT(BP_GET_LEVEL(bp) == 0); |
1455 | bp->blk_fill = 1; | |
1456 | } | |
b128c09f BB |
1457 | } |
1458 | } | |
1459 | ||
428870ff BB |
1460 | static void |
1461 | dmu_sync_late_arrival_ready(zio_t *zio) | |
1462 | { | |
1463 | dmu_sync_ready(zio, NULL, zio->io_private); | |
1464 | } | |
1465 | ||
34dc7c2f BB |
1466 | /* ARGSUSED */ |
1467 | static void | |
1468 | dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) | |
1469 | { | |
428870ff BB |
1470 | dmu_sync_arg_t *dsa = varg; |
1471 | dbuf_dirty_record_t *dr = dsa->dsa_dr; | |
34dc7c2f | 1472 | dmu_buf_impl_t *db = dr->dr_dbuf; |
34dc7c2f | 1473 | |
34dc7c2f BB |
1474 | mutex_enter(&db->db_mtx); |
1475 | ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); | |
428870ff BB |
1476 | if (zio->io_error == 0) { |
1477 | dr->dt.dl.dr_overridden_by = *zio->io_bp; | |
1478 | dr->dt.dl.dr_override_state = DR_OVERRIDDEN; | |
1479 | dr->dt.dl.dr_copies = zio->io_prop.zp_copies; | |
1480 | if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by)) | |
1481 | BP_ZERO(&dr->dt.dl.dr_overridden_by); | |
1482 | } else { | |
1483 | dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; | |
1484 | } | |
34dc7c2f BB |
1485 | cv_broadcast(&db->db_changed); |
1486 | mutex_exit(&db->db_mtx); | |
1487 | ||
428870ff | 1488 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); |
34dc7c2f | 1489 | |
428870ff BB |
1490 | kmem_free(dsa, sizeof (*dsa)); |
1491 | } | |
1492 | ||
1493 | static void | |
1494 | dmu_sync_late_arrival_done(zio_t *zio) | |
1495 | { | |
1496 | blkptr_t *bp = zio->io_bp; | |
1497 | dmu_sync_arg_t *dsa = zio->io_private; | |
1498 | ||
1499 | if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { | |
1500 | ASSERT(zio->io_bp->blk_birth == zio->io_txg); | |
1501 | ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); | |
1502 | zio_free(zio->io_spa, zio->io_txg, zio->io_bp); | |
1503 | } | |
1504 | ||
1505 | dmu_tx_commit(dsa->dsa_tx); | |
1506 | ||
1507 | dsa->dsa_done(dsa->dsa_zgd, zio->io_error); | |
1508 | ||
1509 | kmem_free(dsa, sizeof (*dsa)); | |
1510 | } | |
1511 | ||
1512 | static int | |
1513 | dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, | |
1514 | zio_prop_t *zp, zbookmark_t *zb) | |
1515 | { | |
1516 | dmu_sync_arg_t *dsa; | |
1517 | dmu_tx_t *tx; | |
1518 | ||
1519 | tx = dmu_tx_create(os); | |
1520 | dmu_tx_hold_space(tx, zgd->zgd_db->db_size); | |
1521 | if (dmu_tx_assign(tx, TXG_WAIT) != 0) { | |
1522 | dmu_tx_abort(tx); | |
1523 | return (EIO); /* Make zl_get_data do txg_waited_synced() */ | |
1524 | } | |
1525 | ||
b8d06fca | 1526 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_PUSHPAGE); |
428870ff BB |
1527 | dsa->dsa_dr = NULL; |
1528 | dsa->dsa_done = done; | |
1529 | dsa->dsa_zgd = zgd; | |
1530 | dsa->dsa_tx = tx; | |
1531 | ||
1532 | zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, | |
1533 | zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, | |
1534 | dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa, | |
920dd524 | 1535 | ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL | ZIO_FLAG_FASTWRITE, zb)); |
428870ff BB |
1536 | |
1537 | return (0); | |
34dc7c2f BB |
1538 | } |
1539 | ||
1540 | /* | |
1541 | * Intent log support: sync the block associated with db to disk. | |
1542 | * N.B. and XXX: the caller is responsible for making sure that the | |
1543 | * data isn't changing while dmu_sync() is writing it. | |
1544 | * | |
1545 | * Return values: | |
1546 | * | |
1547 | * EEXIST: this txg has already been synced, so there's nothing to to. | |
1548 | * The caller should not log the write. | |
1549 | * | |
1550 | * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. | |
1551 | * The caller should not log the write. | |
1552 | * | |
1553 | * EALREADY: this block is already in the process of being synced. | |
1554 | * The caller should track its progress (somehow). | |
1555 | * | |
428870ff BB |
1556 | * EIO: could not do the I/O. |
1557 | * The caller should do a txg_wait_synced(). | |
34dc7c2f | 1558 | * |
428870ff BB |
1559 | * 0: the I/O has been initiated. |
1560 | * The caller should log this blkptr in the done callback. | |
1561 | * It is possible that the I/O will fail, in which case | |
1562 | * the error will be reported to the done callback and | |
1563 | * propagated to pio from zio_done(). | |
34dc7c2f BB |
1564 | */ |
1565 | int | |
428870ff | 1566 | dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) |
34dc7c2f | 1567 | { |
428870ff BB |
1568 | blkptr_t *bp = zgd->zgd_bp; |
1569 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; | |
1570 | objset_t *os = db->db_objset; | |
1571 | dsl_dataset_t *ds = os->os_dsl_dataset; | |
34dc7c2f | 1572 | dbuf_dirty_record_t *dr; |
428870ff | 1573 | dmu_sync_arg_t *dsa; |
34dc7c2f | 1574 | zbookmark_t zb; |
428870ff | 1575 | zio_prop_t zp; |
572e2857 | 1576 | dnode_t *dn; |
34dc7c2f | 1577 | |
428870ff | 1578 | ASSERT(pio != NULL); |
34dc7c2f BB |
1579 | ASSERT(BP_IS_HOLE(bp)); |
1580 | ASSERT(txg != 0); | |
1581 | ||
428870ff BB |
1582 | SET_BOOKMARK(&zb, ds->ds_object, |
1583 | db->db.db_object, db->db_level, db->db_blkid); | |
1584 | ||
572e2857 BB |
1585 | DB_DNODE_ENTER(db); |
1586 | dn = DB_DNODE(db); | |
1587 | dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); | |
1588 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
1589 | |
1590 | /* | |
428870ff | 1591 | * If we're frozen (running ziltest), we always need to generate a bp. |
34dc7c2f | 1592 | */ |
428870ff BB |
1593 | if (txg > spa_freeze_txg(os->os_spa)) |
1594 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1595 | |
1596 | /* | |
428870ff BB |
1597 | * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() |
1598 | * and us. If we determine that this txg is not yet syncing, | |
1599 | * but it begins to sync a moment later, that's OK because the | |
1600 | * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. | |
34dc7c2f | 1601 | */ |
428870ff BB |
1602 | mutex_enter(&db->db_mtx); |
1603 | ||
1604 | if (txg <= spa_last_synced_txg(os->os_spa)) { | |
34dc7c2f | 1605 | /* |
428870ff | 1606 | * This txg has already synced. There's nothing to do. |
34dc7c2f | 1607 | */ |
428870ff | 1608 | mutex_exit(&db->db_mtx); |
34dc7c2f BB |
1609 | return (EEXIST); |
1610 | } | |
1611 | ||
428870ff BB |
1612 | if (txg <= spa_syncing_txg(os->os_spa)) { |
1613 | /* | |
1614 | * This txg is currently syncing, so we can't mess with | |
1615 | * the dirty record anymore; just write a new log block. | |
1616 | */ | |
1617 | mutex_exit(&db->db_mtx); | |
1618 | return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); | |
34dc7c2f BB |
1619 | } |
1620 | ||
1621 | dr = db->db_last_dirty; | |
428870ff | 1622 | while (dr && dr->dr_txg != txg) |
34dc7c2f | 1623 | dr = dr->dr_next; |
428870ff BB |
1624 | |
1625 | if (dr == NULL) { | |
34dc7c2f | 1626 | /* |
428870ff | 1627 | * There's no dr for this dbuf, so it must have been freed. |
34dc7c2f BB |
1628 | * There's no need to log writes to freed blocks, so we're done. |
1629 | */ | |
1630 | mutex_exit(&db->db_mtx); | |
34dc7c2f BB |
1631 | return (ENOENT); |
1632 | } | |
1633 | ||
1634 | ASSERT(dr->dr_txg == txg); | |
428870ff BB |
1635 | if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || |
1636 | dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { | |
34dc7c2f | 1637 | /* |
428870ff BB |
1638 | * We have already issued a sync write for this buffer, |
1639 | * or this buffer has already been synced. It could not | |
34dc7c2f BB |
1640 | * have been dirtied since, or we would have cleared the state. |
1641 | */ | |
34dc7c2f | 1642 | mutex_exit(&db->db_mtx); |
428870ff | 1643 | return (EALREADY); |
34dc7c2f BB |
1644 | } |
1645 | ||
428870ff | 1646 | ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); |
34dc7c2f | 1647 | dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; |
34dc7c2f | 1648 | mutex_exit(&db->db_mtx); |
34dc7c2f | 1649 | |
b8d06fca | 1650 | dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_PUSHPAGE); |
428870ff BB |
1651 | dsa->dsa_dr = dr; |
1652 | dsa->dsa_done = done; | |
1653 | dsa->dsa_zgd = zgd; | |
1654 | dsa->dsa_tx = NULL; | |
b128c09f | 1655 | |
428870ff | 1656 | zio_nowait(arc_write(pio, os->os_spa, txg, |
3a17a7a9 SK |
1657 | bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), |
1658 | DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready, dmu_sync_done, | |
1659 | dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL | ZIO_FLAG_FASTWRITE, &zb)); | |
b128c09f | 1660 | |
428870ff | 1661 | return (0); |
34dc7c2f BB |
1662 | } |
1663 | ||
1664 | int | |
1665 | dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, | |
1666 | dmu_tx_t *tx) | |
1667 | { | |
1668 | dnode_t *dn; | |
1669 | int err; | |
1670 | ||
428870ff | 1671 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1672 | if (err) |
1673 | return (err); | |
1674 | err = dnode_set_blksz(dn, size, ibs, tx); | |
1675 | dnode_rele(dn, FTAG); | |
1676 | return (err); | |
1677 | } | |
1678 | ||
1679 | void | |
1680 | dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, | |
1681 | dmu_tx_t *tx) | |
1682 | { | |
1683 | dnode_t *dn; | |
1684 | ||
1685 | /* XXX assumes dnode_hold will not get an i/o error */ | |
428870ff | 1686 | (void) dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1687 | ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS); |
1688 | dn->dn_checksum = checksum; | |
1689 | dnode_setdirty(dn, tx); | |
1690 | dnode_rele(dn, FTAG); | |
1691 | } | |
1692 | ||
1693 | void | |
1694 | dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, | |
1695 | dmu_tx_t *tx) | |
1696 | { | |
1697 | dnode_t *dn; | |
1698 | ||
1699 | /* XXX assumes dnode_hold will not get an i/o error */ | |
428870ff | 1700 | (void) dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1701 | ASSERT(compress < ZIO_COMPRESS_FUNCTIONS); |
1702 | dn->dn_compress = compress; | |
1703 | dnode_setdirty(dn, tx); | |
1704 | dnode_rele(dn, FTAG); | |
1705 | } | |
1706 | ||
428870ff BB |
1707 | int zfs_mdcomp_disable = 0; |
1708 | ||
1709 | void | |
1710 | dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) | |
1711 | { | |
1712 | dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; | |
9ae529ec | 1713 | boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || |
572e2857 | 1714 | (wp & WP_SPILL)); |
428870ff BB |
1715 | enum zio_checksum checksum = os->os_checksum; |
1716 | enum zio_compress compress = os->os_compress; | |
1717 | enum zio_checksum dedup_checksum = os->os_dedup_checksum; | |
1718 | boolean_t dedup; | |
1719 | boolean_t dedup_verify = os->os_dedup_verify; | |
1720 | int copies = os->os_copies; | |
1721 | ||
1722 | /* | |
1723 | * Determine checksum setting. | |
1724 | */ | |
1725 | if (ismd) { | |
1726 | /* | |
1727 | * Metadata always gets checksummed. If the data | |
1728 | * checksum is multi-bit correctable, and it's not a | |
1729 | * ZBT-style checksum, then it's suitable for metadata | |
1730 | * as well. Otherwise, the metadata checksum defaults | |
1731 | * to fletcher4. | |
1732 | */ | |
1733 | if (zio_checksum_table[checksum].ci_correctable < 1 || | |
1734 | zio_checksum_table[checksum].ci_eck) | |
1735 | checksum = ZIO_CHECKSUM_FLETCHER_4; | |
1736 | } else { | |
1737 | checksum = zio_checksum_select(dn->dn_checksum, checksum); | |
1738 | } | |
1739 | ||
1740 | /* | |
1741 | * Determine compression setting. | |
1742 | */ | |
1743 | if (ismd) { | |
1744 | /* | |
1745 | * XXX -- we should design a compression algorithm | |
1746 | * that specializes in arrays of bps. | |
1747 | */ | |
1748 | compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY : | |
1749 | ZIO_COMPRESS_LZJB; | |
1750 | } else { | |
1751 | compress = zio_compress_select(dn->dn_compress, compress); | |
1752 | } | |
1753 | ||
1754 | /* | |
1755 | * Determine dedup setting. If we are in dmu_sync(), we won't | |
1756 | * actually dedup now because that's all done in syncing context; | |
1757 | * but we do want to use the dedup checkum. If the checksum is not | |
1758 | * strong enough to ensure unique signatures, force dedup_verify. | |
1759 | */ | |
1760 | dedup = (!ismd && dedup_checksum != ZIO_CHECKSUM_OFF); | |
1761 | if (dedup) { | |
1762 | checksum = dedup_checksum; | |
1763 | if (!zio_checksum_table[checksum].ci_dedup) | |
1764 | dedup_verify = 1; | |
1765 | } | |
1766 | ||
1767 | if (wp & WP_DMU_SYNC) | |
1768 | dedup = 0; | |
1769 | ||
1770 | if (wp & WP_NOFILL) { | |
1771 | ASSERT(!ismd && level == 0); | |
1772 | checksum = ZIO_CHECKSUM_OFF; | |
1773 | compress = ZIO_COMPRESS_OFF; | |
1774 | dedup = B_FALSE; | |
1775 | } | |
1776 | ||
1777 | zp->zp_checksum = checksum; | |
1778 | zp->zp_compress = compress; | |
1779 | zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; | |
1780 | zp->zp_level = level; | |
1781 | zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa)); | |
1782 | zp->zp_dedup = dedup; | |
1783 | zp->zp_dedup_verify = dedup && dedup_verify; | |
1784 | } | |
1785 | ||
34dc7c2f BB |
1786 | int |
1787 | dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) | |
1788 | { | |
1789 | dnode_t *dn; | |
1790 | int i, err; | |
1791 | ||
428870ff | 1792 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1793 | if (err) |
1794 | return (err); | |
1795 | /* | |
1796 | * Sync any current changes before | |
1797 | * we go trundling through the block pointers. | |
1798 | */ | |
1799 | for (i = 0; i < TXG_SIZE; i++) { | |
1800 | if (list_link_active(&dn->dn_dirty_link[i])) | |
1801 | break; | |
1802 | } | |
1803 | if (i != TXG_SIZE) { | |
1804 | dnode_rele(dn, FTAG); | |
1805 | txg_wait_synced(dmu_objset_pool(os), 0); | |
428870ff | 1806 | err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1807 | if (err) |
1808 | return (err); | |
1809 | } | |
1810 | ||
b128c09f | 1811 | err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); |
34dc7c2f BB |
1812 | dnode_rele(dn, FTAG); |
1813 | ||
1814 | return (err); | |
1815 | } | |
1816 | ||
1817 | void | |
1818 | dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) | |
1819 | { | |
428870ff | 1820 | dnode_phys_t *dnp; |
d6320ddb | 1821 | int i; |
428870ff | 1822 | |
34dc7c2f BB |
1823 | rw_enter(&dn->dn_struct_rwlock, RW_READER); |
1824 | mutex_enter(&dn->dn_mtx); | |
1825 | ||
428870ff BB |
1826 | dnp = dn->dn_phys; |
1827 | ||
34dc7c2f BB |
1828 | doi->doi_data_block_size = dn->dn_datablksz; |
1829 | doi->doi_metadata_block_size = dn->dn_indblkshift ? | |
1830 | 1ULL << dn->dn_indblkshift : 0; | |
428870ff BB |
1831 | doi->doi_type = dn->dn_type; |
1832 | doi->doi_bonus_type = dn->dn_bonustype; | |
1833 | doi->doi_bonus_size = dn->dn_bonuslen; | |
34dc7c2f BB |
1834 | doi->doi_indirection = dn->dn_nlevels; |
1835 | doi->doi_checksum = dn->dn_checksum; | |
1836 | doi->doi_compress = dn->dn_compress; | |
428870ff BB |
1837 | doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; |
1838 | doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz; | |
1839 | doi->doi_fill_count = 0; | |
d6320ddb | 1840 | for (i = 0; i < dnp->dn_nblkptr; i++) |
428870ff | 1841 | doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill; |
34dc7c2f BB |
1842 | |
1843 | mutex_exit(&dn->dn_mtx); | |
1844 | rw_exit(&dn->dn_struct_rwlock); | |
1845 | } | |
1846 | ||
1847 | /* | |
1848 | * Get information on a DMU object. | |
1849 | * If doi is NULL, just indicates whether the object exists. | |
1850 | */ | |
1851 | int | |
1852 | dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) | |
1853 | { | |
1854 | dnode_t *dn; | |
428870ff | 1855 | int err = dnode_hold(os, object, FTAG, &dn); |
34dc7c2f BB |
1856 | |
1857 | if (err) | |
1858 | return (err); | |
1859 | ||
1860 | if (doi != NULL) | |
1861 | dmu_object_info_from_dnode(dn, doi); | |
1862 | ||
1863 | dnode_rele(dn, FTAG); | |
1864 | return (0); | |
1865 | } | |
1866 | ||
1867 | /* | |
1868 | * As above, but faster; can be used when you have a held dbuf in hand. | |
1869 | */ | |
1870 | void | |
572e2857 | 1871 | dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) |
34dc7c2f | 1872 | { |
572e2857 BB |
1873 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
1874 | ||
1875 | DB_DNODE_ENTER(db); | |
1876 | dmu_object_info_from_dnode(DB_DNODE(db), doi); | |
1877 | DB_DNODE_EXIT(db); | |
34dc7c2f BB |
1878 | } |
1879 | ||
1880 | /* | |
1881 | * Faster still when you only care about the size. | |
1882 | * This is specifically optimized for zfs_getattr(). | |
1883 | */ | |
1884 | void | |
572e2857 BB |
1885 | dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, |
1886 | u_longlong_t *nblk512) | |
34dc7c2f | 1887 | { |
572e2857 BB |
1888 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; |
1889 | dnode_t *dn; | |
1890 | ||
1891 | DB_DNODE_ENTER(db); | |
1892 | dn = DB_DNODE(db); | |
34dc7c2f BB |
1893 | |
1894 | *blksize = dn->dn_datablksz; | |
1895 | /* add 1 for dnode space */ | |
1896 | *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> | |
1897 | SPA_MINBLOCKSHIFT) + 1; | |
572e2857 | 1898 | DB_DNODE_EXIT(db); |
34dc7c2f BB |
1899 | } |
1900 | ||
1901 | void | |
1902 | byteswap_uint64_array(void *vbuf, size_t size) | |
1903 | { | |
1904 | uint64_t *buf = vbuf; | |
1905 | size_t count = size >> 3; | |
1906 | int i; | |
1907 | ||
1908 | ASSERT((size & 7) == 0); | |
1909 | ||
1910 | for (i = 0; i < count; i++) | |
1911 | buf[i] = BSWAP_64(buf[i]); | |
1912 | } | |
1913 | ||
1914 | void | |
1915 | byteswap_uint32_array(void *vbuf, size_t size) | |
1916 | { | |
1917 | uint32_t *buf = vbuf; | |
1918 | size_t count = size >> 2; | |
1919 | int i; | |
1920 | ||
1921 | ASSERT((size & 3) == 0); | |
1922 | ||
1923 | for (i = 0; i < count; i++) | |
1924 | buf[i] = BSWAP_32(buf[i]); | |
1925 | } | |
1926 | ||
1927 | void | |
1928 | byteswap_uint16_array(void *vbuf, size_t size) | |
1929 | { | |
1930 | uint16_t *buf = vbuf; | |
1931 | size_t count = size >> 1; | |
1932 | int i; | |
1933 | ||
1934 | ASSERT((size & 1) == 0); | |
1935 | ||
1936 | for (i = 0; i < count; i++) | |
1937 | buf[i] = BSWAP_16(buf[i]); | |
1938 | } | |
1939 | ||
1940 | /* ARGSUSED */ | |
1941 | void | |
1942 | byteswap_uint8_array(void *vbuf, size_t size) | |
1943 | { | |
1944 | } | |
1945 | ||
1946 | void | |
1947 | dmu_init(void) | |
1948 | { | |
428870ff | 1949 | zfs_dbgmsg_init(); |
572e2857 BB |
1950 | sa_cache_init(); |
1951 | xuio_stat_init(); | |
1952 | dmu_objset_init(); | |
34dc7c2f | 1953 | dnode_init(); |
572e2857 | 1954 | dbuf_init(); |
428870ff | 1955 | zfetch_init(); |
570827e1 | 1956 | dmu_tx_init(); |
34dc7c2f | 1957 | l2arc_init(); |
29809a6c | 1958 | arc_init(); |
34dc7c2f BB |
1959 | } |
1960 | ||
1961 | void | |
1962 | dmu_fini(void) | |
1963 | { | |
1964 | arc_fini(); | |
29809a6c | 1965 | l2arc_fini(); |
570827e1 | 1966 | dmu_tx_fini(); |
428870ff | 1967 | zfetch_fini(); |
34dc7c2f | 1968 | dbuf_fini(); |
572e2857 BB |
1969 | dnode_fini(); |
1970 | dmu_objset_fini(); | |
428870ff BB |
1971 | xuio_stat_fini(); |
1972 | sa_cache_fini(); | |
1973 | zfs_dbgmsg_fini(); | |
34dc7c2f | 1974 | } |
c28b2279 BB |
1975 | |
1976 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
1977 | EXPORT_SYMBOL(dmu_bonus_hold); | |
a473d90c AZ |
1978 | EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus); |
1979 | EXPORT_SYMBOL(dmu_buf_rele_array); | |
57b650b8 | 1980 | EXPORT_SYMBOL(dmu_prefetch); |
c28b2279 | 1981 | EXPORT_SYMBOL(dmu_free_range); |
57b650b8 BB |
1982 | EXPORT_SYMBOL(dmu_free_long_range); |
1983 | EXPORT_SYMBOL(dmu_free_object); | |
c28b2279 BB |
1984 | EXPORT_SYMBOL(dmu_read); |
1985 | EXPORT_SYMBOL(dmu_write); | |
57b650b8 | 1986 | EXPORT_SYMBOL(dmu_prealloc); |
c28b2279 BB |
1987 | EXPORT_SYMBOL(dmu_object_info); |
1988 | EXPORT_SYMBOL(dmu_object_info_from_dnode); | |
1989 | EXPORT_SYMBOL(dmu_object_info_from_db); | |
1990 | EXPORT_SYMBOL(dmu_object_size_from_db); | |
c28b2279 BB |
1991 | EXPORT_SYMBOL(dmu_object_set_blocksize); |
1992 | EXPORT_SYMBOL(dmu_object_set_checksum); | |
1993 | EXPORT_SYMBOL(dmu_object_set_compress); | |
57b650b8 BB |
1994 | EXPORT_SYMBOL(dmu_write_policy); |
1995 | EXPORT_SYMBOL(dmu_sync); | |
b10c77f7 BB |
1996 | EXPORT_SYMBOL(dmu_request_arcbuf); |
1997 | EXPORT_SYMBOL(dmu_return_arcbuf); | |
1998 | EXPORT_SYMBOL(dmu_assign_arcbuf); | |
1999 | EXPORT_SYMBOL(dmu_buf_hold); | |
c28b2279 | 2000 | EXPORT_SYMBOL(dmu_ot); |
afec56b4 BB |
2001 | |
2002 | module_param(zfs_mdcomp_disable, int, 0644); | |
2003 | MODULE_PARM_DESC(zfs_mdcomp_disable, "Disable meta data compression"); | |
c28b2279 | 2004 | #endif |