]>
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. |
9bd274dd | 23 | * Copyright (c) 2012, 2014 by Delphix. All rights reserved. |
0c66c32d | 24 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. |
34dc7c2f BB |
25 | */ |
26 | ||
34dc7c2f BB |
27 | #include <sys/zfs_context.h> |
28 | #include <sys/dbuf.h> | |
29 | #include <sys/dnode.h> | |
30 | #include <sys/dmu.h> | |
31 | #include <sys/dmu_impl.h> | |
32 | #include <sys/dmu_tx.h> | |
33 | #include <sys/dmu_objset.h> | |
34 | #include <sys/dsl_dir.h> | |
35 | #include <sys/dsl_dataset.h> | |
36 | #include <sys/spa.h> | |
37 | #include <sys/zio.h> | |
38 | #include <sys/dmu_zfetch.h> | |
9bd274dd | 39 | #include <sys/range_tree.h> |
49ee64e5 | 40 | #include <sys/trace_dnode.h> |
34dc7c2f BB |
41 | |
42 | static kmem_cache_t *dnode_cache; | |
572e2857 BB |
43 | /* |
44 | * Define DNODE_STATS to turn on statistic gathering. By default, it is only | |
45 | * turned on when DEBUG is also defined. | |
46 | */ | |
47 | #ifdef DEBUG | |
48 | #define DNODE_STATS | |
49 | #endif /* DEBUG */ | |
50 | ||
51 | #ifdef DNODE_STATS | |
52 | #define DNODE_STAT_ADD(stat) ((stat)++) | |
53 | #else | |
54 | #define DNODE_STAT_ADD(stat) /* nothing */ | |
55 | #endif /* DNODE_STATS */ | |
34dc7c2f | 56 | |
1fde1e37 | 57 | ASSERTV(static dnode_phys_t dnode_phys_zero); |
34dc7c2f BB |
58 | |
59 | int zfs_default_bs = SPA_MINBLOCKSHIFT; | |
60 | int zfs_default_ibs = DN_MAX_INDBLKSHIFT; | |
61 | ||
5ac1241a | 62 | #ifdef _KERNEL |
572e2857 | 63 | static kmem_cbrc_t dnode_move(void *, void *, size_t, void *); |
5ac1241a | 64 | #endif /* _KERNEL */ |
572e2857 | 65 | |
8951cb8d AR |
66 | static int |
67 | dbuf_compare(const void *x1, const void *x2) | |
68 | { | |
69 | const dmu_buf_impl_t *d1 = x1; | |
70 | const dmu_buf_impl_t *d2 = x2; | |
71 | ||
72 | if (d1->db_level < d2->db_level) { | |
73 | return (-1); | |
9925c28c AR |
74 | } |
75 | if (d1->db_level > d2->db_level) { | |
8951cb8d AR |
76 | return (1); |
77 | } | |
78 | ||
79 | if (d1->db_blkid < d2->db_blkid) { | |
80 | return (-1); | |
9925c28c AR |
81 | } |
82 | if (d1->db_blkid > d2->db_blkid) { | |
8951cb8d AR |
83 | return (1); |
84 | } | |
85 | ||
9925c28c | 86 | if (d1->db_state < d2->db_state) { |
8951cb8d | 87 | return (-1); |
9925c28c AR |
88 | } |
89 | if (d1->db_state > d2->db_state) { | |
90 | return (1); | |
91 | } | |
92 | ||
93 | ASSERT3S(d1->db_state, !=, DB_SEARCH); | |
94 | ASSERT3S(d2->db_state, !=, DB_SEARCH); | |
95 | ||
96 | if ((uintptr_t)d1 < (uintptr_t)d2) { | |
97 | return (-1); | |
98 | } | |
99 | if ((uintptr_t)d1 > (uintptr_t)d2) { | |
8951cb8d | 100 | return (1); |
8951cb8d | 101 | } |
9925c28c | 102 | return (0); |
8951cb8d AR |
103 | } |
104 | ||
34dc7c2f BB |
105 | /* ARGSUSED */ |
106 | static int | |
107 | dnode_cons(void *arg, void *unused, int kmflag) | |
108 | { | |
34dc7c2f | 109 | dnode_t *dn = arg; |
572e2857 | 110 | int i; |
34dc7c2f BB |
111 | |
112 | rw_init(&dn->dn_struct_rwlock, NULL, RW_DEFAULT, NULL); | |
113 | mutex_init(&dn->dn_mtx, NULL, MUTEX_DEFAULT, NULL); | |
114 | mutex_init(&dn->dn_dbufs_mtx, NULL, MUTEX_DEFAULT, NULL); | |
fb5f0bc8 BB |
115 | cv_init(&dn->dn_notxholds, NULL, CV_DEFAULT, NULL); |
116 | ||
13fe0198 MA |
117 | /* |
118 | * Every dbuf has a reference, and dropping a tracked reference is | |
119 | * O(number of references), so don't track dn_holds. | |
120 | */ | |
121 | refcount_create_untracked(&dn->dn_holds); | |
34dc7c2f | 122 | refcount_create(&dn->dn_tx_holds); |
572e2857 BB |
123 | list_link_init(&dn->dn_link); |
124 | ||
125 | bzero(&dn->dn_next_nblkptr[0], sizeof (dn->dn_next_nblkptr)); | |
126 | bzero(&dn->dn_next_nlevels[0], sizeof (dn->dn_next_nlevels)); | |
127 | bzero(&dn->dn_next_indblkshift[0], sizeof (dn->dn_next_indblkshift)); | |
128 | bzero(&dn->dn_next_bonustype[0], sizeof (dn->dn_next_bonustype)); | |
129 | bzero(&dn->dn_rm_spillblk[0], sizeof (dn->dn_rm_spillblk)); | |
130 | bzero(&dn->dn_next_bonuslen[0], sizeof (dn->dn_next_bonuslen)); | |
131 | bzero(&dn->dn_next_blksz[0], sizeof (dn->dn_next_blksz)); | |
34dc7c2f BB |
132 | |
133 | for (i = 0; i < TXG_SIZE; i++) { | |
572e2857 | 134 | list_link_init(&dn->dn_dirty_link[i]); |
9bd274dd | 135 | dn->dn_free_ranges[i] = NULL; |
34dc7c2f BB |
136 | list_create(&dn->dn_dirty_records[i], |
137 | sizeof (dbuf_dirty_record_t), | |
138 | offsetof(dbuf_dirty_record_t, dr_dirty_node)); | |
139 | } | |
140 | ||
572e2857 BB |
141 | dn->dn_allocated_txg = 0; |
142 | dn->dn_free_txg = 0; | |
143 | dn->dn_assigned_txg = 0; | |
144 | dn->dn_dirtyctx = 0; | |
145 | dn->dn_dirtyctx_firstset = NULL; | |
146 | dn->dn_bonus = NULL; | |
147 | dn->dn_have_spill = B_FALSE; | |
148 | dn->dn_zio = NULL; | |
149 | dn->dn_oldused = 0; | |
150 | dn->dn_oldflags = 0; | |
151 | dn->dn_olduid = 0; | |
152 | dn->dn_oldgid = 0; | |
153 | dn->dn_newuid = 0; | |
154 | dn->dn_newgid = 0; | |
155 | dn->dn_id_flags = 0; | |
156 | ||
157 | dn->dn_dbufs_count = 0; | |
b663a23d | 158 | dn->dn_unlisted_l0_blkid = 0; |
8951cb8d | 159 | avl_create(&dn->dn_dbufs, dbuf_compare, sizeof (dmu_buf_impl_t), |
34dc7c2f BB |
160 | offsetof(dmu_buf_impl_t, db_link)); |
161 | ||
572e2857 | 162 | dn->dn_moved = 0; |
34dc7c2f BB |
163 | return (0); |
164 | } | |
165 | ||
166 | /* ARGSUSED */ | |
167 | static void | |
168 | dnode_dest(void *arg, void *unused) | |
169 | { | |
170 | int i; | |
171 | dnode_t *dn = arg; | |
172 | ||
173 | rw_destroy(&dn->dn_struct_rwlock); | |
174 | mutex_destroy(&dn->dn_mtx); | |
175 | mutex_destroy(&dn->dn_dbufs_mtx); | |
fb5f0bc8 | 176 | cv_destroy(&dn->dn_notxholds); |
34dc7c2f BB |
177 | refcount_destroy(&dn->dn_holds); |
178 | refcount_destroy(&dn->dn_tx_holds); | |
572e2857 | 179 | ASSERT(!list_link_active(&dn->dn_link)); |
34dc7c2f BB |
180 | |
181 | for (i = 0; i < TXG_SIZE; i++) { | |
572e2857 | 182 | ASSERT(!list_link_active(&dn->dn_dirty_link[i])); |
9bd274dd | 183 | ASSERT3P(dn->dn_free_ranges[i], ==, NULL); |
34dc7c2f | 184 | list_destroy(&dn->dn_dirty_records[i]); |
c99c9001 MS |
185 | ASSERT0(dn->dn_next_nblkptr[i]); |
186 | ASSERT0(dn->dn_next_nlevels[i]); | |
187 | ASSERT0(dn->dn_next_indblkshift[i]); | |
188 | ASSERT0(dn->dn_next_bonustype[i]); | |
189 | ASSERT0(dn->dn_rm_spillblk[i]); | |
190 | ASSERT0(dn->dn_next_bonuslen[i]); | |
191 | ASSERT0(dn->dn_next_blksz[i]); | |
34dc7c2f BB |
192 | } |
193 | ||
c99c9001 MS |
194 | ASSERT0(dn->dn_allocated_txg); |
195 | ASSERT0(dn->dn_free_txg); | |
196 | ASSERT0(dn->dn_assigned_txg); | |
197 | ASSERT0(dn->dn_dirtyctx); | |
572e2857 BB |
198 | ASSERT3P(dn->dn_dirtyctx_firstset, ==, NULL); |
199 | ASSERT3P(dn->dn_bonus, ==, NULL); | |
200 | ASSERT(!dn->dn_have_spill); | |
201 | ASSERT3P(dn->dn_zio, ==, NULL); | |
c99c9001 MS |
202 | ASSERT0(dn->dn_oldused); |
203 | ASSERT0(dn->dn_oldflags); | |
204 | ASSERT0(dn->dn_olduid); | |
205 | ASSERT0(dn->dn_oldgid); | |
206 | ASSERT0(dn->dn_newuid); | |
207 | ASSERT0(dn->dn_newgid); | |
208 | ASSERT0(dn->dn_id_flags); | |
209 | ||
210 | ASSERT0(dn->dn_dbufs_count); | |
b663a23d | 211 | ASSERT0(dn->dn_unlisted_l0_blkid); |
8951cb8d | 212 | avl_destroy(&dn->dn_dbufs); |
34dc7c2f BB |
213 | } |
214 | ||
215 | void | |
216 | dnode_init(void) | |
217 | { | |
572e2857 | 218 | ASSERT(dnode_cache == NULL); |
ae6ba3db | 219 | dnode_cache = kmem_cache_create("dnode_t", sizeof (dnode_t), |
6795a698 | 220 | 0, dnode_cons, dnode_dest, NULL, NULL, NULL, 0); |
572e2857 | 221 | kmem_cache_set_move(dnode_cache, dnode_move); |
34dc7c2f BB |
222 | } |
223 | ||
224 | void | |
225 | dnode_fini(void) | |
226 | { | |
227 | kmem_cache_destroy(dnode_cache); | |
572e2857 | 228 | dnode_cache = NULL; |
34dc7c2f BB |
229 | } |
230 | ||
231 | ||
232 | #ifdef ZFS_DEBUG | |
233 | void | |
234 | dnode_verify(dnode_t *dn) | |
235 | { | |
236 | int drop_struct_lock = FALSE; | |
237 | ||
238 | ASSERT(dn->dn_phys); | |
239 | ASSERT(dn->dn_objset); | |
572e2857 | 240 | ASSERT(dn->dn_handle->dnh_dnode == dn); |
34dc7c2f | 241 | |
9ae529ec | 242 | ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type)); |
34dc7c2f BB |
243 | |
244 | if (!(zfs_flags & ZFS_DEBUG_DNODE_VERIFY)) | |
245 | return; | |
246 | ||
247 | if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { | |
248 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
249 | drop_struct_lock = TRUE; | |
250 | } | |
251 | if (dn->dn_phys->dn_type != DMU_OT_NONE || dn->dn_allocated_txg != 0) { | |
252 | int i; | |
34dc7c2f BB |
253 | ASSERT3U(dn->dn_indblkshift, <=, SPA_MAXBLOCKSHIFT); |
254 | if (dn->dn_datablkshift) { | |
255 | ASSERT3U(dn->dn_datablkshift, >=, SPA_MINBLOCKSHIFT); | |
256 | ASSERT3U(dn->dn_datablkshift, <=, SPA_MAXBLOCKSHIFT); | |
257 | ASSERT3U(1<<dn->dn_datablkshift, ==, dn->dn_datablksz); | |
258 | } | |
259 | ASSERT3U(dn->dn_nlevels, <=, 30); | |
9ae529ec | 260 | ASSERT(DMU_OT_IS_VALID(dn->dn_type)); |
34dc7c2f BB |
261 | ASSERT3U(dn->dn_nblkptr, >=, 1); |
262 | ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); | |
263 | ASSERT3U(dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); | |
264 | ASSERT3U(dn->dn_datablksz, ==, | |
265 | dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
266 | ASSERT3U(ISP2(dn->dn_datablksz), ==, dn->dn_datablkshift != 0); | |
267 | ASSERT3U((dn->dn_nblkptr - 1) * sizeof (blkptr_t) + | |
268 | dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); | |
269 | for (i = 0; i < TXG_SIZE; i++) { | |
270 | ASSERT3U(dn->dn_next_nlevels[i], <=, dn->dn_nlevels); | |
271 | } | |
272 | } | |
273 | if (dn->dn_phys->dn_type != DMU_OT_NONE) | |
274 | ASSERT3U(dn->dn_phys->dn_nlevels, <=, dn->dn_nlevels); | |
9babb374 | 275 | ASSERT(DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_dbuf != NULL); |
34dc7c2f BB |
276 | if (dn->dn_dbuf != NULL) { |
277 | ASSERT3P(dn->dn_phys, ==, | |
278 | (dnode_phys_t *)dn->dn_dbuf->db.db_data + | |
279 | (dn->dn_object % (dn->dn_dbuf->db.db_size >> DNODE_SHIFT))); | |
280 | } | |
281 | if (drop_struct_lock) | |
282 | rw_exit(&dn->dn_struct_rwlock); | |
283 | } | |
284 | #endif | |
285 | ||
286 | void | |
287 | dnode_byteswap(dnode_phys_t *dnp) | |
288 | { | |
289 | uint64_t *buf64 = (void*)&dnp->dn_blkptr; | |
290 | int i; | |
291 | ||
292 | if (dnp->dn_type == DMU_OT_NONE) { | |
293 | bzero(dnp, sizeof (dnode_phys_t)); | |
294 | return; | |
295 | } | |
296 | ||
297 | dnp->dn_datablkszsec = BSWAP_16(dnp->dn_datablkszsec); | |
298 | dnp->dn_bonuslen = BSWAP_16(dnp->dn_bonuslen); | |
299 | dnp->dn_maxblkid = BSWAP_64(dnp->dn_maxblkid); | |
300 | dnp->dn_used = BSWAP_64(dnp->dn_used); | |
301 | ||
302 | /* | |
303 | * dn_nblkptr is only one byte, so it's OK to read it in either | |
304 | * byte order. We can't read dn_bouslen. | |
305 | */ | |
306 | ASSERT(dnp->dn_indblkshift <= SPA_MAXBLOCKSHIFT); | |
307 | ASSERT(dnp->dn_nblkptr <= DN_MAX_NBLKPTR); | |
308 | for (i = 0; i < dnp->dn_nblkptr * sizeof (blkptr_t)/8; i++) | |
309 | buf64[i] = BSWAP_64(buf64[i]); | |
310 | ||
311 | /* | |
312 | * OK to check dn_bonuslen for zero, because it won't matter if | |
313 | * we have the wrong byte order. This is necessary because the | |
314 | * dnode dnode is smaller than a regular dnode. | |
315 | */ | |
316 | if (dnp->dn_bonuslen != 0) { | |
317 | /* | |
318 | * Note that the bonus length calculated here may be | |
319 | * longer than the actual bonus buffer. This is because | |
320 | * we always put the bonus buffer after the last block | |
321 | * pointer (instead of packing it against the end of the | |
322 | * dnode buffer). | |
323 | */ | |
324 | int off = (dnp->dn_nblkptr-1) * sizeof (blkptr_t); | |
325 | size_t len = DN_MAX_BONUSLEN - off; | |
9ae529ec CS |
326 | dmu_object_byteswap_t byteswap; |
327 | ASSERT(DMU_OT_IS_VALID(dnp->dn_bonustype)); | |
328 | byteswap = DMU_OT_BYTESWAP(dnp->dn_bonustype); | |
329 | dmu_ot_byteswap[byteswap].ob_func(dnp->dn_bonus + off, len); | |
34dc7c2f | 330 | } |
428870ff BB |
331 | |
332 | /* Swap SPILL block if we have one */ | |
333 | if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) | |
334 | byteswap_uint64_array(&dnp->dn_spill, sizeof (blkptr_t)); | |
335 | ||
34dc7c2f BB |
336 | } |
337 | ||
338 | void | |
339 | dnode_buf_byteswap(void *vbuf, size_t size) | |
340 | { | |
341 | dnode_phys_t *buf = vbuf; | |
342 | int i; | |
343 | ||
344 | ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT)); | |
345 | ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0); | |
346 | ||
347 | size >>= DNODE_SHIFT; | |
348 | for (i = 0; i < size; i++) { | |
349 | dnode_byteswap(buf); | |
350 | buf++; | |
351 | } | |
352 | } | |
353 | ||
34dc7c2f BB |
354 | void |
355 | dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx) | |
356 | { | |
357 | ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); | |
358 | ||
359 | dnode_setdirty(dn, tx); | |
360 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
361 | ASSERT3U(newsize, <=, DN_MAX_BONUSLEN - | |
362 | (dn->dn_nblkptr-1) * sizeof (blkptr_t)); | |
363 | dn->dn_bonuslen = newsize; | |
364 | if (newsize == 0) | |
365 | dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = DN_ZERO_BONUSLEN; | |
366 | else | |
367 | dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; | |
368 | rw_exit(&dn->dn_struct_rwlock); | |
369 | } | |
370 | ||
428870ff BB |
371 | void |
372 | dnode_setbonus_type(dnode_t *dn, dmu_object_type_t newtype, dmu_tx_t *tx) | |
373 | { | |
374 | ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); | |
375 | dnode_setdirty(dn, tx); | |
376 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
377 | dn->dn_bonustype = newtype; | |
378 | dn->dn_next_bonustype[tx->tx_txg & TXG_MASK] = dn->dn_bonustype; | |
379 | rw_exit(&dn->dn_struct_rwlock); | |
380 | } | |
381 | ||
382 | void | |
383 | dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx) | |
384 | { | |
385 | ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); | |
386 | ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); | |
387 | dnode_setdirty(dn, tx); | |
388 | dn->dn_rm_spillblk[tx->tx_txg&TXG_MASK] = DN_KILL_SPILLBLK; | |
389 | dn->dn_have_spill = B_FALSE; | |
390 | } | |
391 | ||
34dc7c2f BB |
392 | static void |
393 | dnode_setdblksz(dnode_t *dn, int size) | |
394 | { | |
c99c9001 | 395 | ASSERT0(P2PHASE(size, SPA_MINBLOCKSIZE)); |
34dc7c2f BB |
396 | ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); |
397 | ASSERT3U(size, >=, SPA_MINBLOCKSIZE); | |
398 | ASSERT3U(size >> SPA_MINBLOCKSHIFT, <, | |
399 | 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8)); | |
400 | dn->dn_datablksz = size; | |
401 | dn->dn_datablkszsec = size >> SPA_MINBLOCKSHIFT; | |
9bd274dd | 402 | dn->dn_datablkshift = ISP2(size) ? highbit64(size - 1) : 0; |
34dc7c2f BB |
403 | } |
404 | ||
405 | static dnode_t * | |
428870ff | 406 | dnode_create(objset_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, |
572e2857 | 407 | uint64_t object, dnode_handle_t *dnh) |
34dc7c2f | 408 | { |
0c66c32d | 409 | dnode_t *dn; |
34dc7c2f | 410 | |
0c66c32d | 411 | dn = kmem_cache_alloc(dnode_cache, KM_SLEEP); |
572e2857 BB |
412 | ASSERT(!POINTER_IS_VALID(dn->dn_objset)); |
413 | dn->dn_moved = 0; | |
414 | ||
415 | /* | |
416 | * Defer setting dn_objset until the dnode is ready to be a candidate | |
417 | * for the dnode_move() callback. | |
418 | */ | |
34dc7c2f BB |
419 | dn->dn_object = object; |
420 | dn->dn_dbuf = db; | |
572e2857 | 421 | dn->dn_handle = dnh; |
34dc7c2f BB |
422 | dn->dn_phys = dnp; |
423 | ||
572e2857 | 424 | if (dnp->dn_datablkszsec) { |
34dc7c2f | 425 | dnode_setdblksz(dn, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); |
572e2857 BB |
426 | } else { |
427 | dn->dn_datablksz = 0; | |
428 | dn->dn_datablkszsec = 0; | |
429 | dn->dn_datablkshift = 0; | |
430 | } | |
34dc7c2f BB |
431 | dn->dn_indblkshift = dnp->dn_indblkshift; |
432 | dn->dn_nlevels = dnp->dn_nlevels; | |
433 | dn->dn_type = dnp->dn_type; | |
434 | dn->dn_nblkptr = dnp->dn_nblkptr; | |
435 | dn->dn_checksum = dnp->dn_checksum; | |
436 | dn->dn_compress = dnp->dn_compress; | |
437 | dn->dn_bonustype = dnp->dn_bonustype; | |
438 | dn->dn_bonuslen = dnp->dn_bonuslen; | |
439 | dn->dn_maxblkid = dnp->dn_maxblkid; | |
428870ff BB |
440 | dn->dn_have_spill = ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0); |
441 | dn->dn_id_flags = 0; | |
34dc7c2f BB |
442 | |
443 | dmu_zfetch_init(&dn->dn_zfetch, dn); | |
444 | ||
9ae529ec | 445 | ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type)); |
572e2857 | 446 | |
34dc7c2f | 447 | mutex_enter(&os->os_lock); |
0c66c32d JG |
448 | if (dnh->dnh_dnode != NULL) { |
449 | /* Lost the allocation race. */ | |
450 | mutex_exit(&os->os_lock); | |
451 | kmem_cache_free(dnode_cache, dn); | |
452 | return (dnh->dnh_dnode); | |
453 | } | |
454 | ||
455 | /* | |
456 | * Exclude special dnodes from os_dnodes so an empty os_dnodes | |
457 | * signifies that the special dnodes have no references from | |
458 | * their children (the entries in os_dnodes). This allows | |
459 | * dnode_destroy() to easily determine if the last child has | |
460 | * been removed and then complete eviction of the objset. | |
461 | */ | |
462 | if (!DMU_OBJECT_IS_SPECIAL(object)) | |
463 | list_insert_head(&os->os_dnodes, dn); | |
572e2857 | 464 | membar_producer(); |
0c66c32d | 465 | |
572e2857 | 466 | /* |
0c66c32d JG |
467 | * Everything else must be valid before assigning dn_objset |
468 | * makes the dnode eligible for dnode_move(). | |
572e2857 BB |
469 | */ |
470 | dn->dn_objset = os; | |
0c66c32d JG |
471 | |
472 | dnh->dnh_dnode = dn; | |
34dc7c2f BB |
473 | mutex_exit(&os->os_lock); |
474 | ||
d164b209 | 475 | arc_space_consume(sizeof (dnode_t), ARC_SPACE_OTHER); |
34dc7c2f BB |
476 | return (dn); |
477 | } | |
478 | ||
572e2857 BB |
479 | /* |
480 | * Caller must be holding the dnode handle, which is released upon return. | |
481 | */ | |
34dc7c2f BB |
482 | static void |
483 | dnode_destroy(dnode_t *dn) | |
484 | { | |
428870ff | 485 | objset_t *os = dn->dn_objset; |
0c66c32d | 486 | boolean_t complete_os_eviction = B_FALSE; |
34dc7c2f | 487 | |
428870ff | 488 | ASSERT((dn->dn_id_flags & DN_ID_NEW_EXIST) == 0); |
34dc7c2f BB |
489 | |
490 | mutex_enter(&os->os_lock); | |
572e2857 | 491 | POINTER_INVALIDATE(&dn->dn_objset); |
0c66c32d JG |
492 | if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { |
493 | list_remove(&os->os_dnodes, dn); | |
494 | complete_os_eviction = | |
495 | list_is_empty(&os->os_dnodes) && | |
496 | list_link_active(&os->os_evicting_node); | |
497 | } | |
34dc7c2f BB |
498 | mutex_exit(&os->os_lock); |
499 | ||
572e2857 BB |
500 | /* the dnode can no longer move, so we can release the handle */ |
501 | zrl_remove(&dn->dn_handle->dnh_zrlock); | |
502 | ||
503 | dn->dn_allocated_txg = 0; | |
504 | dn->dn_free_txg = 0; | |
505 | dn->dn_assigned_txg = 0; | |
506 | ||
507 | dn->dn_dirtyctx = 0; | |
508 | if (dn->dn_dirtyctx_firstset != NULL) { | |
34dc7c2f BB |
509 | kmem_free(dn->dn_dirtyctx_firstset, 1); |
510 | dn->dn_dirtyctx_firstset = NULL; | |
511 | } | |
572e2857 | 512 | if (dn->dn_bonus != NULL) { |
34dc7c2f BB |
513 | mutex_enter(&dn->dn_bonus->db_mtx); |
514 | dbuf_evict(dn->dn_bonus); | |
515 | dn->dn_bonus = NULL; | |
516 | } | |
572e2857 BB |
517 | dn->dn_zio = NULL; |
518 | ||
519 | dn->dn_have_spill = B_FALSE; | |
520 | dn->dn_oldused = 0; | |
521 | dn->dn_oldflags = 0; | |
522 | dn->dn_olduid = 0; | |
523 | dn->dn_oldgid = 0; | |
524 | dn->dn_newuid = 0; | |
525 | dn->dn_newgid = 0; | |
526 | dn->dn_id_flags = 0; | |
b663a23d | 527 | dn->dn_unlisted_l0_blkid = 0; |
572e2857 BB |
528 | |
529 | dmu_zfetch_rele(&dn->dn_zfetch); | |
34dc7c2f | 530 | kmem_cache_free(dnode_cache, dn); |
d164b209 | 531 | arc_space_return(sizeof (dnode_t), ARC_SPACE_OTHER); |
0c66c32d JG |
532 | |
533 | if (complete_os_eviction) | |
534 | dmu_objset_evict_done(os); | |
34dc7c2f BB |
535 | } |
536 | ||
537 | void | |
538 | dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, | |
539 | dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) | |
540 | { | |
541 | int i; | |
542 | ||
f1512ee6 MA |
543 | ASSERT3U(blocksize, <=, |
544 | spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); | |
34dc7c2f BB |
545 | if (blocksize == 0) |
546 | blocksize = 1 << zfs_default_bs; | |
34dc7c2f BB |
547 | else |
548 | blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); | |
549 | ||
550 | if (ibs == 0) | |
551 | ibs = zfs_default_ibs; | |
552 | ||
553 | ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT); | |
554 | ||
555 | dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn->dn_objset, | |
556 | dn->dn_object, tx->tx_txg, blocksize, ibs); | |
557 | ||
558 | ASSERT(dn->dn_type == DMU_OT_NONE); | |
559 | ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); | |
560 | ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); | |
561 | ASSERT(ot != DMU_OT_NONE); | |
9ae529ec | 562 | ASSERT(DMU_OT_IS_VALID(ot)); |
34dc7c2f | 563 | ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || |
428870ff | 564 | (bonustype == DMU_OT_SA && bonuslen == 0) || |
34dc7c2f | 565 | (bonustype != DMU_OT_NONE && bonuslen != 0)); |
9ae529ec | 566 | ASSERT(DMU_OT_IS_VALID(bonustype)); |
34dc7c2f BB |
567 | ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); |
568 | ASSERT(dn->dn_type == DMU_OT_NONE); | |
c99c9001 MS |
569 | ASSERT0(dn->dn_maxblkid); |
570 | ASSERT0(dn->dn_allocated_txg); | |
571 | ASSERT0(dn->dn_assigned_txg); | |
34dc7c2f BB |
572 | ASSERT(refcount_is_zero(&dn->dn_tx_holds)); |
573 | ASSERT3U(refcount_count(&dn->dn_holds), <=, 1); | |
8951cb8d | 574 | ASSERT(avl_is_empty(&dn->dn_dbufs)); |
34dc7c2f BB |
575 | |
576 | for (i = 0; i < TXG_SIZE; i++) { | |
c99c9001 MS |
577 | ASSERT0(dn->dn_next_nblkptr[i]); |
578 | ASSERT0(dn->dn_next_nlevels[i]); | |
579 | ASSERT0(dn->dn_next_indblkshift[i]); | |
580 | ASSERT0(dn->dn_next_bonuslen[i]); | |
581 | ASSERT0(dn->dn_next_bonustype[i]); | |
582 | ASSERT0(dn->dn_rm_spillblk[i]); | |
583 | ASSERT0(dn->dn_next_blksz[i]); | |
34dc7c2f BB |
584 | ASSERT(!list_link_active(&dn->dn_dirty_link[i])); |
585 | ASSERT3P(list_head(&dn->dn_dirty_records[i]), ==, NULL); | |
9bd274dd | 586 | ASSERT3P(dn->dn_free_ranges[i], ==, NULL); |
34dc7c2f BB |
587 | } |
588 | ||
589 | dn->dn_type = ot; | |
590 | dnode_setdblksz(dn, blocksize); | |
591 | dn->dn_indblkshift = ibs; | |
592 | dn->dn_nlevels = 1; | |
428870ff BB |
593 | if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */ |
594 | dn->dn_nblkptr = 1; | |
595 | else | |
596 | dn->dn_nblkptr = 1 + | |
597 | ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); | |
34dc7c2f BB |
598 | dn->dn_bonustype = bonustype; |
599 | dn->dn_bonuslen = bonuslen; | |
600 | dn->dn_checksum = ZIO_CHECKSUM_INHERIT; | |
601 | dn->dn_compress = ZIO_COMPRESS_INHERIT; | |
602 | dn->dn_dirtyctx = 0; | |
603 | ||
604 | dn->dn_free_txg = 0; | |
605 | if (dn->dn_dirtyctx_firstset) { | |
606 | kmem_free(dn->dn_dirtyctx_firstset, 1); | |
607 | dn->dn_dirtyctx_firstset = NULL; | |
608 | } | |
609 | ||
610 | dn->dn_allocated_txg = tx->tx_txg; | |
428870ff | 611 | dn->dn_id_flags = 0; |
34dc7c2f BB |
612 | |
613 | dnode_setdirty(dn, tx); | |
614 | dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs; | |
615 | dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; | |
428870ff | 616 | dn->dn_next_bonustype[tx->tx_txg & TXG_MASK] = dn->dn_bonustype; |
34dc7c2f BB |
617 | dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz; |
618 | } | |
619 | ||
620 | void | |
621 | dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, | |
622 | dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) | |
623 | { | |
9babb374 | 624 | int nblkptr; |
34dc7c2f BB |
625 | |
626 | ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); | |
f1512ee6 MA |
627 | ASSERT3U(blocksize, <=, |
628 | spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); | |
c99c9001 | 629 | ASSERT0(blocksize % SPA_MINBLOCKSIZE); |
34dc7c2f BB |
630 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); |
631 | ASSERT(tx->tx_txg != 0); | |
632 | ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || | |
428870ff BB |
633 | (bonustype != DMU_OT_NONE && bonuslen != 0) || |
634 | (bonustype == DMU_OT_SA && bonuslen == 0)); | |
9ae529ec | 635 | ASSERT(DMU_OT_IS_VALID(bonustype)); |
34dc7c2f BB |
636 | ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); |
637 | ||
34dc7c2f BB |
638 | /* clean up any unreferenced dbufs */ |
639 | dnode_evict_dbufs(dn); | |
d164b209 | 640 | |
428870ff BB |
641 | dn->dn_id_flags = 0; |
642 | ||
34dc7c2f | 643 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); |
34dc7c2f | 644 | dnode_setdirty(dn, tx); |
9babb374 BB |
645 | if (dn->dn_datablksz != blocksize) { |
646 | /* change blocksize */ | |
647 | ASSERT(dn->dn_maxblkid == 0 && | |
648 | (BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || | |
649 | dnode_block_freed(dn, 0))); | |
650 | dnode_setdblksz(dn, blocksize); | |
651 | dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize; | |
652 | } | |
653 | if (dn->dn_bonuslen != bonuslen) | |
654 | dn->dn_next_bonuslen[tx->tx_txg&TXG_MASK] = bonuslen; | |
428870ff BB |
655 | |
656 | if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */ | |
657 | nblkptr = 1; | |
658 | else | |
659 | nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); | |
660 | if (dn->dn_bonustype != bonustype) | |
661 | dn->dn_next_bonustype[tx->tx_txg&TXG_MASK] = bonustype; | |
d164b209 BB |
662 | if (dn->dn_nblkptr != nblkptr) |
663 | dn->dn_next_nblkptr[tx->tx_txg&TXG_MASK] = nblkptr; | |
428870ff BB |
664 | if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { |
665 | dbuf_rm_spill(dn, tx); | |
666 | dnode_rm_spill(dn, tx); | |
667 | } | |
34dc7c2f | 668 | rw_exit(&dn->dn_struct_rwlock); |
34dc7c2f BB |
669 | |
670 | /* change type */ | |
671 | dn->dn_type = ot; | |
672 | ||
673 | /* change bonus size and type */ | |
674 | mutex_enter(&dn->dn_mtx); | |
34dc7c2f BB |
675 | dn->dn_bonustype = bonustype; |
676 | dn->dn_bonuslen = bonuslen; | |
d164b209 | 677 | dn->dn_nblkptr = nblkptr; |
34dc7c2f BB |
678 | dn->dn_checksum = ZIO_CHECKSUM_INHERIT; |
679 | dn->dn_compress = ZIO_COMPRESS_INHERIT; | |
680 | ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); | |
681 | ||
d164b209 BB |
682 | /* fix up the bonus db_size */ |
683 | if (dn->dn_bonus) { | |
34dc7c2f BB |
684 | dn->dn_bonus->db.db_size = |
685 | DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t); | |
686 | ASSERT(dn->dn_bonuslen <= dn->dn_bonus->db.db_size); | |
687 | } | |
688 | ||
689 | dn->dn_allocated_txg = tx->tx_txg; | |
690 | mutex_exit(&dn->dn_mtx); | |
691 | } | |
692 | ||
5ac1241a | 693 | #ifdef _KERNEL |
572e2857 BB |
694 | #ifdef DNODE_STATS |
695 | static struct { | |
696 | uint64_t dms_dnode_invalid; | |
697 | uint64_t dms_dnode_recheck1; | |
698 | uint64_t dms_dnode_recheck2; | |
699 | uint64_t dms_dnode_special; | |
700 | uint64_t dms_dnode_handle; | |
701 | uint64_t dms_dnode_rwlock; | |
702 | uint64_t dms_dnode_active; | |
703 | } dnode_move_stats; | |
704 | #endif /* DNODE_STATS */ | |
705 | ||
706 | static void | |
707 | dnode_move_impl(dnode_t *odn, dnode_t *ndn) | |
708 | { | |
709 | int i; | |
710 | ||
711 | ASSERT(!RW_LOCK_HELD(&odn->dn_struct_rwlock)); | |
712 | ASSERT(MUTEX_NOT_HELD(&odn->dn_mtx)); | |
713 | ASSERT(MUTEX_NOT_HELD(&odn->dn_dbufs_mtx)); | |
714 | ASSERT(!RW_LOCK_HELD(&odn->dn_zfetch.zf_rwlock)); | |
715 | ||
716 | /* Copy fields. */ | |
717 | ndn->dn_objset = odn->dn_objset; | |
718 | ndn->dn_object = odn->dn_object; | |
719 | ndn->dn_dbuf = odn->dn_dbuf; | |
720 | ndn->dn_handle = odn->dn_handle; | |
721 | ndn->dn_phys = odn->dn_phys; | |
722 | ndn->dn_type = odn->dn_type; | |
723 | ndn->dn_bonuslen = odn->dn_bonuslen; | |
724 | ndn->dn_bonustype = odn->dn_bonustype; | |
725 | ndn->dn_nblkptr = odn->dn_nblkptr; | |
726 | ndn->dn_checksum = odn->dn_checksum; | |
727 | ndn->dn_compress = odn->dn_compress; | |
728 | ndn->dn_nlevels = odn->dn_nlevels; | |
729 | ndn->dn_indblkshift = odn->dn_indblkshift; | |
730 | ndn->dn_datablkshift = odn->dn_datablkshift; | |
731 | ndn->dn_datablkszsec = odn->dn_datablkszsec; | |
732 | ndn->dn_datablksz = odn->dn_datablksz; | |
733 | ndn->dn_maxblkid = odn->dn_maxblkid; | |
734 | bcopy(&odn->dn_next_nblkptr[0], &ndn->dn_next_nblkptr[0], | |
735 | sizeof (odn->dn_next_nblkptr)); | |
736 | bcopy(&odn->dn_next_nlevels[0], &ndn->dn_next_nlevels[0], | |
737 | sizeof (odn->dn_next_nlevels)); | |
738 | bcopy(&odn->dn_next_indblkshift[0], &ndn->dn_next_indblkshift[0], | |
739 | sizeof (odn->dn_next_indblkshift)); | |
740 | bcopy(&odn->dn_next_bonustype[0], &ndn->dn_next_bonustype[0], | |
741 | sizeof (odn->dn_next_bonustype)); | |
742 | bcopy(&odn->dn_rm_spillblk[0], &ndn->dn_rm_spillblk[0], | |
743 | sizeof (odn->dn_rm_spillblk)); | |
744 | bcopy(&odn->dn_next_bonuslen[0], &ndn->dn_next_bonuslen[0], | |
745 | sizeof (odn->dn_next_bonuslen)); | |
746 | bcopy(&odn->dn_next_blksz[0], &ndn->dn_next_blksz[0], | |
747 | sizeof (odn->dn_next_blksz)); | |
748 | for (i = 0; i < TXG_SIZE; i++) { | |
749 | list_move_tail(&ndn->dn_dirty_records[i], | |
750 | &odn->dn_dirty_records[i]); | |
751 | } | |
9bd274dd MA |
752 | bcopy(&odn->dn_free_ranges[0], &ndn->dn_free_ranges[0], |
753 | sizeof (odn->dn_free_ranges)); | |
572e2857 BB |
754 | ndn->dn_allocated_txg = odn->dn_allocated_txg; |
755 | ndn->dn_free_txg = odn->dn_free_txg; | |
756 | ndn->dn_assigned_txg = odn->dn_assigned_txg; | |
757 | ndn->dn_dirtyctx = odn->dn_dirtyctx; | |
758 | ndn->dn_dirtyctx_firstset = odn->dn_dirtyctx_firstset; | |
759 | ASSERT(refcount_count(&odn->dn_tx_holds) == 0); | |
760 | refcount_transfer(&ndn->dn_holds, &odn->dn_holds); | |
8951cb8d AR |
761 | ASSERT(avl_is_empty(&ndn->dn_dbufs)); |
762 | avl_swap(&ndn->dn_dbufs, &odn->dn_dbufs); | |
572e2857 | 763 | ndn->dn_dbufs_count = odn->dn_dbufs_count; |
b663a23d | 764 | ndn->dn_unlisted_l0_blkid = odn->dn_unlisted_l0_blkid; |
572e2857 BB |
765 | ndn->dn_bonus = odn->dn_bonus; |
766 | ndn->dn_have_spill = odn->dn_have_spill; | |
767 | ndn->dn_zio = odn->dn_zio; | |
768 | ndn->dn_oldused = odn->dn_oldused; | |
769 | ndn->dn_oldflags = odn->dn_oldflags; | |
770 | ndn->dn_olduid = odn->dn_olduid; | |
771 | ndn->dn_oldgid = odn->dn_oldgid; | |
772 | ndn->dn_newuid = odn->dn_newuid; | |
773 | ndn->dn_newgid = odn->dn_newgid; | |
774 | ndn->dn_id_flags = odn->dn_id_flags; | |
775 | dmu_zfetch_init(&ndn->dn_zfetch, NULL); | |
776 | list_move_tail(&ndn->dn_zfetch.zf_stream, &odn->dn_zfetch.zf_stream); | |
777 | ndn->dn_zfetch.zf_dnode = odn->dn_zfetch.zf_dnode; | |
778 | ndn->dn_zfetch.zf_stream_cnt = odn->dn_zfetch.zf_stream_cnt; | |
779 | ndn->dn_zfetch.zf_alloc_fail = odn->dn_zfetch.zf_alloc_fail; | |
780 | ||
781 | /* | |
782 | * Update back pointers. Updating the handle fixes the back pointer of | |
783 | * every descendant dbuf as well as the bonus dbuf. | |
784 | */ | |
785 | ASSERT(ndn->dn_handle->dnh_dnode == odn); | |
786 | ndn->dn_handle->dnh_dnode = ndn; | |
787 | if (ndn->dn_zfetch.zf_dnode == odn) { | |
788 | ndn->dn_zfetch.zf_dnode = ndn; | |
789 | } | |
790 | ||
791 | /* | |
792 | * Invalidate the original dnode by clearing all of its back pointers. | |
793 | */ | |
794 | odn->dn_dbuf = NULL; | |
795 | odn->dn_handle = NULL; | |
8951cb8d | 796 | avl_create(&odn->dn_dbufs, dbuf_compare, sizeof (dmu_buf_impl_t), |
572e2857 BB |
797 | offsetof(dmu_buf_impl_t, db_link)); |
798 | odn->dn_dbufs_count = 0; | |
b663a23d | 799 | odn->dn_unlisted_l0_blkid = 0; |
572e2857 BB |
800 | odn->dn_bonus = NULL; |
801 | odn->dn_zfetch.zf_dnode = NULL; | |
802 | ||
803 | /* | |
804 | * Set the low bit of the objset pointer to ensure that dnode_move() | |
805 | * recognizes the dnode as invalid in any subsequent callback. | |
806 | */ | |
807 | POINTER_INVALIDATE(&odn->dn_objset); | |
808 | ||
809 | /* | |
810 | * Satisfy the destructor. | |
811 | */ | |
812 | for (i = 0; i < TXG_SIZE; i++) { | |
813 | list_create(&odn->dn_dirty_records[i], | |
814 | sizeof (dbuf_dirty_record_t), | |
815 | offsetof(dbuf_dirty_record_t, dr_dirty_node)); | |
9bd274dd | 816 | odn->dn_free_ranges[i] = NULL; |
572e2857 BB |
817 | odn->dn_next_nlevels[i] = 0; |
818 | odn->dn_next_indblkshift[i] = 0; | |
819 | odn->dn_next_bonustype[i] = 0; | |
820 | odn->dn_rm_spillblk[i] = 0; | |
821 | odn->dn_next_bonuslen[i] = 0; | |
822 | odn->dn_next_blksz[i] = 0; | |
823 | } | |
824 | odn->dn_allocated_txg = 0; | |
825 | odn->dn_free_txg = 0; | |
826 | odn->dn_assigned_txg = 0; | |
827 | odn->dn_dirtyctx = 0; | |
828 | odn->dn_dirtyctx_firstset = NULL; | |
829 | odn->dn_have_spill = B_FALSE; | |
830 | odn->dn_zio = NULL; | |
831 | odn->dn_oldused = 0; | |
832 | odn->dn_oldflags = 0; | |
833 | odn->dn_olduid = 0; | |
834 | odn->dn_oldgid = 0; | |
835 | odn->dn_newuid = 0; | |
836 | odn->dn_newgid = 0; | |
837 | odn->dn_id_flags = 0; | |
838 | ||
839 | /* | |
840 | * Mark the dnode. | |
841 | */ | |
842 | ndn->dn_moved = 1; | |
843 | odn->dn_moved = (uint8_t)-1; | |
844 | } | |
845 | ||
572e2857 BB |
846 | /*ARGSUSED*/ |
847 | static kmem_cbrc_t | |
848 | dnode_move(void *buf, void *newbuf, size_t size, void *arg) | |
849 | { | |
850 | dnode_t *odn = buf, *ndn = newbuf; | |
851 | objset_t *os; | |
852 | int64_t refcount; | |
853 | uint32_t dbufs; | |
854 | ||
855 | /* | |
856 | * The dnode is on the objset's list of known dnodes if the objset | |
857 | * pointer is valid. We set the low bit of the objset pointer when | |
858 | * freeing the dnode to invalidate it, and the memory patterns written | |
859 | * by kmem (baddcafe and deadbeef) set at least one of the two low bits. | |
860 | * A newly created dnode sets the objset pointer last of all to indicate | |
861 | * that the dnode is known and in a valid state to be moved by this | |
862 | * function. | |
863 | */ | |
864 | os = odn->dn_objset; | |
865 | if (!POINTER_IS_VALID(os)) { | |
866 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_invalid); | |
867 | return (KMEM_CBRC_DONT_KNOW); | |
868 | } | |
869 | ||
870 | /* | |
871 | * Ensure that the objset does not go away during the move. | |
872 | */ | |
873 | rw_enter(&os_lock, RW_WRITER); | |
874 | if (os != odn->dn_objset) { | |
875 | rw_exit(&os_lock); | |
876 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_recheck1); | |
877 | return (KMEM_CBRC_DONT_KNOW); | |
878 | } | |
879 | ||
880 | /* | |
881 | * If the dnode is still valid, then so is the objset. We know that no | |
882 | * valid objset can be freed while we hold os_lock, so we can safely | |
883 | * ensure that the objset remains in use. | |
884 | */ | |
885 | mutex_enter(&os->os_lock); | |
886 | ||
887 | /* | |
888 | * Recheck the objset pointer in case the dnode was removed just before | |
889 | * acquiring the lock. | |
890 | */ | |
891 | if (os != odn->dn_objset) { | |
892 | mutex_exit(&os->os_lock); | |
893 | rw_exit(&os_lock); | |
894 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_recheck2); | |
895 | return (KMEM_CBRC_DONT_KNOW); | |
896 | } | |
897 | ||
898 | /* | |
899 | * At this point we know that as long as we hold os->os_lock, the dnode | |
900 | * cannot be freed and fields within the dnode can be safely accessed. | |
901 | * The objset listing this dnode cannot go away as long as this dnode is | |
902 | * on its list. | |
903 | */ | |
904 | rw_exit(&os_lock); | |
905 | if (DMU_OBJECT_IS_SPECIAL(odn->dn_object)) { | |
906 | mutex_exit(&os->os_lock); | |
907 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_special); | |
908 | return (KMEM_CBRC_NO); | |
909 | } | |
910 | ASSERT(odn->dn_dbuf != NULL); /* only "special" dnodes have no parent */ | |
911 | ||
912 | /* | |
913 | * Lock the dnode handle to prevent the dnode from obtaining any new | |
914 | * holds. This also prevents the descendant dbufs and the bonus dbuf | |
915 | * from accessing the dnode, so that we can discount their holds. The | |
916 | * handle is safe to access because we know that while the dnode cannot | |
917 | * go away, neither can its handle. Once we hold dnh_zrlock, we can | |
918 | * safely move any dnode referenced only by dbufs. | |
919 | */ | |
920 | if (!zrl_tryenter(&odn->dn_handle->dnh_zrlock)) { | |
921 | mutex_exit(&os->os_lock); | |
922 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_handle); | |
923 | return (KMEM_CBRC_LATER); | |
924 | } | |
925 | ||
926 | /* | |
927 | * Ensure a consistent view of the dnode's holds and the dnode's dbufs. | |
928 | * We need to guarantee that there is a hold for every dbuf in order to | |
929 | * determine whether the dnode is actively referenced. Falsely matching | |
930 | * a dbuf to an active hold would lead to an unsafe move. It's possible | |
931 | * that a thread already having an active dnode hold is about to add a | |
932 | * dbuf, and we can't compare hold and dbuf counts while the add is in | |
933 | * progress. | |
934 | */ | |
935 | if (!rw_tryenter(&odn->dn_struct_rwlock, RW_WRITER)) { | |
936 | zrl_exit(&odn->dn_handle->dnh_zrlock); | |
937 | mutex_exit(&os->os_lock); | |
938 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_rwlock); | |
939 | return (KMEM_CBRC_LATER); | |
940 | } | |
941 | ||
942 | /* | |
943 | * A dbuf may be removed (evicted) without an active dnode hold. In that | |
944 | * case, the dbuf count is decremented under the handle lock before the | |
945 | * dbuf's hold is released. This order ensures that if we count the hold | |
946 | * after the dbuf is removed but before its hold is released, we will | |
947 | * treat the unmatched hold as active and exit safely. If we count the | |
948 | * hold before the dbuf is removed, the hold is discounted, and the | |
949 | * removal is blocked until the move completes. | |
950 | */ | |
951 | refcount = refcount_count(&odn->dn_holds); | |
952 | ASSERT(refcount >= 0); | |
953 | dbufs = odn->dn_dbufs_count; | |
954 | ||
955 | /* We can't have more dbufs than dnode holds. */ | |
956 | ASSERT3U(dbufs, <=, refcount); | |
957 | DTRACE_PROBE3(dnode__move, dnode_t *, odn, int64_t, refcount, | |
958 | uint32_t, dbufs); | |
959 | ||
960 | if (refcount > dbufs) { | |
961 | rw_exit(&odn->dn_struct_rwlock); | |
962 | zrl_exit(&odn->dn_handle->dnh_zrlock); | |
963 | mutex_exit(&os->os_lock); | |
964 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_active); | |
965 | return (KMEM_CBRC_LATER); | |
966 | } | |
967 | ||
968 | rw_exit(&odn->dn_struct_rwlock); | |
969 | ||
970 | /* | |
971 | * At this point we know that anyone with a hold on the dnode is not | |
972 | * actively referencing it. The dnode is known and in a valid state to | |
973 | * move. We're holding the locks needed to execute the critical section. | |
974 | */ | |
975 | dnode_move_impl(odn, ndn); | |
976 | ||
977 | list_link_replace(&odn->dn_link, &ndn->dn_link); | |
978 | /* If the dnode was safe to move, the refcount cannot have changed. */ | |
979 | ASSERT(refcount == refcount_count(&ndn->dn_holds)); | |
980 | ASSERT(dbufs == ndn->dn_dbufs_count); | |
981 | zrl_exit(&ndn->dn_handle->dnh_zrlock); /* handle has moved */ | |
982 | mutex_exit(&os->os_lock); | |
983 | ||
984 | return (KMEM_CBRC_YES); | |
985 | } | |
986 | #endif /* _KERNEL */ | |
987 | ||
34dc7c2f | 988 | void |
572e2857 | 989 | dnode_special_close(dnode_handle_t *dnh) |
34dc7c2f | 990 | { |
572e2857 BB |
991 | dnode_t *dn = dnh->dnh_dnode; |
992 | ||
34dc7c2f BB |
993 | /* |
994 | * Wait for final references to the dnode to clear. This can | |
995 | * only happen if the arc is asyncronously evicting state that | |
996 | * has a hold on this dnode while we are trying to evict this | |
997 | * dnode. | |
998 | */ | |
999 | while (refcount_count(&dn->dn_holds) > 0) | |
1000 | delay(1); | |
0c66c32d JG |
1001 | ASSERT(dn->dn_dbuf == NULL || |
1002 | dmu_buf_get_user(&dn->dn_dbuf->db) == NULL); | |
572e2857 BB |
1003 | zrl_add(&dnh->dnh_zrlock); |
1004 | dnode_destroy(dn); /* implicit zrl_remove() */ | |
1005 | zrl_destroy(&dnh->dnh_zrlock); | |
1006 | dnh->dnh_dnode = NULL; | |
34dc7c2f BB |
1007 | } |
1008 | ||
0c66c32d | 1009 | void |
572e2857 BB |
1010 | dnode_special_open(objset_t *os, dnode_phys_t *dnp, uint64_t object, |
1011 | dnode_handle_t *dnh) | |
34dc7c2f | 1012 | { |
0c66c32d JG |
1013 | dnode_t *dn; |
1014 | ||
1015 | dn = dnode_create(os, dnp, NULL, object, dnh); | |
572e2857 | 1016 | zrl_init(&dnh->dnh_zrlock); |
34dc7c2f | 1017 | DNODE_VERIFY(dn); |
34dc7c2f BB |
1018 | } |
1019 | ||
1020 | static void | |
0c66c32d | 1021 | dnode_buf_pageout(void *dbu) |
34dc7c2f | 1022 | { |
0c66c32d | 1023 | dnode_children_t *children_dnodes = dbu; |
34dc7c2f | 1024 | int i; |
34dc7c2f | 1025 | |
0c66c32d | 1026 | for (i = 0; i < children_dnodes->dnc_count; i++) { |
572e2857 BB |
1027 | dnode_handle_t *dnh = &children_dnodes->dnc_children[i]; |
1028 | dnode_t *dn; | |
34dc7c2f | 1029 | |
572e2857 BB |
1030 | /* |
1031 | * The dnode handle lock guards against the dnode moving to | |
1032 | * another valid address, so there is no need here to guard | |
1033 | * against changes to or from NULL. | |
1034 | */ | |
1035 | if (dnh->dnh_dnode == NULL) { | |
1036 | zrl_destroy(&dnh->dnh_zrlock); | |
34dc7c2f | 1037 | continue; |
572e2857 BB |
1038 | } |
1039 | ||
1040 | zrl_add(&dnh->dnh_zrlock); | |
1041 | dn = dnh->dnh_dnode; | |
34dc7c2f BB |
1042 | /* |
1043 | * If there are holds on this dnode, then there should | |
1044 | * be holds on the dnode's containing dbuf as well; thus | |
572e2857 | 1045 | * it wouldn't be eligible for eviction and this function |
34dc7c2f BB |
1046 | * would not have been called. |
1047 | */ | |
1048 | ASSERT(refcount_is_zero(&dn->dn_holds)); | |
34dc7c2f BB |
1049 | ASSERT(refcount_is_zero(&dn->dn_tx_holds)); |
1050 | ||
572e2857 BB |
1051 | dnode_destroy(dn); /* implicit zrl_remove() */ |
1052 | zrl_destroy(&dnh->dnh_zrlock); | |
1053 | dnh->dnh_dnode = NULL; | |
34dc7c2f | 1054 | } |
572e2857 | 1055 | kmem_free(children_dnodes, sizeof (dnode_children_t) + |
0c66c32d | 1056 | children_dnodes->dnc_count * sizeof (dnode_handle_t)); |
34dc7c2f BB |
1057 | } |
1058 | ||
1059 | /* | |
1060 | * errors: | |
1061 | * EINVAL - invalid object number. | |
1062 | * EIO - i/o error. | |
1063 | * succeeds even for free dnodes. | |
1064 | */ | |
1065 | int | |
428870ff | 1066 | dnode_hold_impl(objset_t *os, uint64_t object, int flag, |
34dc7c2f BB |
1067 | void *tag, dnode_t **dnp) |
1068 | { | |
1069 | int epb, idx, err; | |
1070 | int drop_struct_lock = FALSE; | |
1071 | int type; | |
1072 | uint64_t blk; | |
1073 | dnode_t *mdn, *dn; | |
1074 | dmu_buf_impl_t *db; | |
572e2857 BB |
1075 | dnode_children_t *children_dnodes; |
1076 | dnode_handle_t *dnh; | |
34dc7c2f | 1077 | |
b128c09f BB |
1078 | /* |
1079 | * If you are holding the spa config lock as writer, you shouldn't | |
428870ff BB |
1080 | * be asking the DMU to do *anything* unless it's the root pool |
1081 | * which may require us to read from the root filesystem while | |
1082 | * holding some (not all) of the locks as writer. | |
b128c09f | 1083 | */ |
428870ff BB |
1084 | ASSERT(spa_config_held(os->os_spa, SCL_ALL, RW_WRITER) == 0 || |
1085 | (spa_is_root(os->os_spa) && | |
572e2857 | 1086 | spa_config_held(os->os_spa, SCL_STATE, RW_WRITER))); |
b128c09f | 1087 | |
9babb374 BB |
1088 | if (object == DMU_USERUSED_OBJECT || object == DMU_GROUPUSED_OBJECT) { |
1089 | dn = (object == DMU_USERUSED_OBJECT) ? | |
572e2857 | 1090 | DMU_USERUSED_DNODE(os) : DMU_GROUPUSED_DNODE(os); |
9babb374 | 1091 | if (dn == NULL) |
2e528b49 | 1092 | return (SET_ERROR(ENOENT)); |
9babb374 BB |
1093 | type = dn->dn_type; |
1094 | if ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) | |
2e528b49 | 1095 | return (SET_ERROR(ENOENT)); |
9babb374 | 1096 | if ((flag & DNODE_MUST_BE_FREE) && type != DMU_OT_NONE) |
2e528b49 | 1097 | return (SET_ERROR(EEXIST)); |
9babb374 BB |
1098 | DNODE_VERIFY(dn); |
1099 | (void) refcount_add(&dn->dn_holds, tag); | |
1100 | *dnp = dn; | |
1101 | return (0); | |
1102 | } | |
1103 | ||
34dc7c2f | 1104 | if (object == 0 || object >= DN_MAX_OBJECT) |
2e528b49 | 1105 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 1106 | |
572e2857 BB |
1107 | mdn = DMU_META_DNODE(os); |
1108 | ASSERT(mdn->dn_object == DMU_META_DNODE_OBJECT); | |
34dc7c2f BB |
1109 | |
1110 | DNODE_VERIFY(mdn); | |
1111 | ||
1112 | if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) { | |
1113 | rw_enter(&mdn->dn_struct_rwlock, RW_READER); | |
1114 | drop_struct_lock = TRUE; | |
1115 | } | |
1116 | ||
1117 | blk = dbuf_whichblock(mdn, object * sizeof (dnode_phys_t)); | |
1118 | ||
1119 | db = dbuf_hold(mdn, blk, FTAG); | |
1120 | if (drop_struct_lock) | |
1121 | rw_exit(&mdn->dn_struct_rwlock); | |
1122 | if (db == NULL) | |
2e528b49 | 1123 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
1124 | err = dbuf_read(db, NULL, DB_RF_CANFAIL); |
1125 | if (err) { | |
1126 | dbuf_rele(db, FTAG); | |
1127 | return (err); | |
1128 | } | |
1129 | ||
1130 | ASSERT3U(db->db.db_size, >=, 1<<DNODE_SHIFT); | |
1131 | epb = db->db.db_size >> DNODE_SHIFT; | |
1132 | ||
1133 | idx = object & (epb-1); | |
1134 | ||
572e2857 | 1135 | ASSERT(DB_DNODE(db)->dn_type == DMU_OT_DNODE); |
34dc7c2f BB |
1136 | children_dnodes = dmu_buf_get_user(&db->db); |
1137 | if (children_dnodes == NULL) { | |
572e2857 BB |
1138 | int i; |
1139 | dnode_children_t *winner; | |
0c66c32d | 1140 | children_dnodes = kmem_zalloc(sizeof (dnode_children_t) + |
5aea3644 | 1141 | epb * sizeof (dnode_handle_t), KM_SLEEP); |
572e2857 BB |
1142 | children_dnodes->dnc_count = epb; |
1143 | dnh = &children_dnodes->dnc_children[0]; | |
1144 | for (i = 0; i < epb; i++) { | |
1145 | zrl_init(&dnh[i].dnh_zrlock); | |
572e2857 | 1146 | } |
0c66c32d JG |
1147 | dmu_buf_init_user(&children_dnodes->dnc_dbu, |
1148 | dnode_buf_pageout, NULL); | |
1149 | winner = dmu_buf_set_user(&db->db, &children_dnodes->dnc_dbu); | |
1150 | if (winner != NULL) { | |
58c4aa00 JL |
1151 | |
1152 | for (i = 0; i < epb; i++) { | |
1153 | zrl_destroy(&dnh[i].dnh_zrlock); | |
1154 | } | |
1155 | ||
572e2857 | 1156 | kmem_free(children_dnodes, sizeof (dnode_children_t) + |
5aea3644 | 1157 | epb * sizeof (dnode_handle_t)); |
34dc7c2f BB |
1158 | children_dnodes = winner; |
1159 | } | |
1160 | } | |
572e2857 | 1161 | ASSERT(children_dnodes->dnc_count == epb); |
34dc7c2f | 1162 | |
572e2857 BB |
1163 | dnh = &children_dnodes->dnc_children[idx]; |
1164 | zrl_add(&dnh->dnh_zrlock); | |
0c66c32d JG |
1165 | dn = dnh->dnh_dnode; |
1166 | if (dn == NULL) { | |
572e2857 | 1167 | dnode_phys_t *phys = (dnode_phys_t *)db->db.db_data+idx; |
34dc7c2f | 1168 | |
572e2857 | 1169 | dn = dnode_create(os, phys, db, object, dnh); |
34dc7c2f BB |
1170 | } |
1171 | ||
1172 | mutex_enter(&dn->dn_mtx); | |
1173 | type = dn->dn_type; | |
1174 | if (dn->dn_free_txg || | |
1175 | ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) || | |
9babb374 | 1176 | ((flag & DNODE_MUST_BE_FREE) && |
428870ff | 1177 | (type != DMU_OT_NONE || !refcount_is_zero(&dn->dn_holds)))) { |
34dc7c2f | 1178 | mutex_exit(&dn->dn_mtx); |
572e2857 | 1179 | zrl_remove(&dnh->dnh_zrlock); |
34dc7c2f BB |
1180 | dbuf_rele(db, FTAG); |
1181 | return (type == DMU_OT_NONE ? ENOENT : EEXIST); | |
1182 | } | |
34dc7c2f | 1183 | if (refcount_add(&dn->dn_holds, tag) == 1) |
572e2857 | 1184 | dbuf_add_ref(db, dnh); |
0c66c32d JG |
1185 | mutex_exit(&dn->dn_mtx); |
1186 | ||
572e2857 BB |
1187 | /* Now we can rely on the hold to prevent the dnode from moving. */ |
1188 | zrl_remove(&dnh->dnh_zrlock); | |
34dc7c2f BB |
1189 | |
1190 | DNODE_VERIFY(dn); | |
1191 | ASSERT3P(dn->dn_dbuf, ==, db); | |
1192 | ASSERT3U(dn->dn_object, ==, object); | |
1193 | dbuf_rele(db, FTAG); | |
1194 | ||
1195 | *dnp = dn; | |
1196 | return (0); | |
1197 | } | |
1198 | ||
1199 | /* | |
1200 | * Return held dnode if the object is allocated, NULL if not. | |
1201 | */ | |
1202 | int | |
428870ff | 1203 | dnode_hold(objset_t *os, uint64_t object, void *tag, dnode_t **dnp) |
34dc7c2f BB |
1204 | { |
1205 | return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp)); | |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * Can only add a reference if there is already at least one | |
1210 | * reference on the dnode. Returns FALSE if unable to add a | |
1211 | * new reference. | |
1212 | */ | |
1213 | boolean_t | |
1214 | dnode_add_ref(dnode_t *dn, void *tag) | |
1215 | { | |
1216 | mutex_enter(&dn->dn_mtx); | |
1217 | if (refcount_is_zero(&dn->dn_holds)) { | |
1218 | mutex_exit(&dn->dn_mtx); | |
1219 | return (FALSE); | |
1220 | } | |
1221 | VERIFY(1 < refcount_add(&dn->dn_holds, tag)); | |
1222 | mutex_exit(&dn->dn_mtx); | |
1223 | return (TRUE); | |
1224 | } | |
1225 | ||
1226 | void | |
1227 | dnode_rele(dnode_t *dn, void *tag) | |
4c7b7eed JG |
1228 | { |
1229 | mutex_enter(&dn->dn_mtx); | |
1230 | dnode_rele_and_unlock(dn, tag); | |
1231 | } | |
1232 | ||
1233 | void | |
1234 | dnode_rele_and_unlock(dnode_t *dn, void *tag) | |
34dc7c2f BB |
1235 | { |
1236 | uint64_t refs; | |
572e2857 BB |
1237 | /* Get while the hold prevents the dnode from moving. */ |
1238 | dmu_buf_impl_t *db = dn->dn_dbuf; | |
1239 | dnode_handle_t *dnh = dn->dn_handle; | |
34dc7c2f | 1240 | |
34dc7c2f BB |
1241 | refs = refcount_remove(&dn->dn_holds, tag); |
1242 | mutex_exit(&dn->dn_mtx); | |
572e2857 BB |
1243 | |
1244 | /* | |
1245 | * It's unsafe to release the last hold on a dnode by dnode_rele() or | |
1246 | * indirectly by dbuf_rele() while relying on the dnode handle to | |
1247 | * prevent the dnode from moving, since releasing the last hold could | |
1248 | * result in the dnode's parent dbuf evicting its dnode handles. For | |
1249 | * that reason anyone calling dnode_rele() or dbuf_rele() without some | |
1250 | * other direct or indirect hold on the dnode must first drop the dnode | |
1251 | * handle. | |
1252 | */ | |
1253 | ASSERT(refs > 0 || dnh->dnh_zrlock.zr_owner != curthread); | |
1254 | ||
34dc7c2f | 1255 | /* NOTE: the DNODE_DNODE does not have a dn_dbuf */ |
572e2857 BB |
1256 | if (refs == 0 && db != NULL) { |
1257 | /* | |
1258 | * Another thread could add a hold to the dnode handle in | |
1259 | * dnode_hold_impl() while holding the parent dbuf. Since the | |
1260 | * hold on the parent dbuf prevents the handle from being | |
1261 | * destroyed, the hold on the handle is OK. We can't yet assert | |
1262 | * that the handle has zero references, but that will be | |
1263 | * asserted anyway when the handle gets destroyed. | |
1264 | */ | |
1265 | dbuf_rele(db, dnh); | |
1266 | } | |
34dc7c2f BB |
1267 | } |
1268 | ||
1269 | void | |
1270 | dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) | |
1271 | { | |
428870ff | 1272 | objset_t *os = dn->dn_objset; |
34dc7c2f BB |
1273 | uint64_t txg = tx->tx_txg; |
1274 | ||
9babb374 BB |
1275 | if (DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { |
1276 | dsl_dataset_dirty(os->os_dsl_dataset, tx); | |
34dc7c2f | 1277 | return; |
9babb374 | 1278 | } |
34dc7c2f BB |
1279 | |
1280 | DNODE_VERIFY(dn); | |
1281 | ||
1282 | #ifdef ZFS_DEBUG | |
1283 | mutex_enter(&dn->dn_mtx); | |
1284 | ASSERT(dn->dn_phys->dn_type || dn->dn_allocated_txg); | |
572e2857 | 1285 | ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); |
34dc7c2f BB |
1286 | mutex_exit(&dn->dn_mtx); |
1287 | #endif | |
1288 | ||
428870ff BB |
1289 | /* |
1290 | * Determine old uid/gid when necessary | |
1291 | */ | |
1292 | dmu_objset_userquota_get_ids(dn, B_TRUE, tx); | |
1293 | ||
34dc7c2f BB |
1294 | mutex_enter(&os->os_lock); |
1295 | ||
1296 | /* | |
1297 | * If we are already marked dirty, we're done. | |
1298 | */ | |
1299 | if (list_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) { | |
1300 | mutex_exit(&os->os_lock); | |
1301 | return; | |
1302 | } | |
1303 | ||
8951cb8d AR |
1304 | ASSERT(!refcount_is_zero(&dn->dn_holds) || |
1305 | !avl_is_empty(&dn->dn_dbufs)); | |
34dc7c2f | 1306 | ASSERT(dn->dn_datablksz != 0); |
c99c9001 MS |
1307 | ASSERT0(dn->dn_next_bonuslen[txg&TXG_MASK]); |
1308 | ASSERT0(dn->dn_next_blksz[txg&TXG_MASK]); | |
1309 | ASSERT0(dn->dn_next_bonustype[txg&TXG_MASK]); | |
34dc7c2f BB |
1310 | |
1311 | dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n", | |
1312 | dn->dn_object, txg); | |
1313 | ||
1314 | if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) { | |
1315 | list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn); | |
1316 | } else { | |
1317 | list_insert_tail(&os->os_dirty_dnodes[txg&TXG_MASK], dn); | |
1318 | } | |
1319 | ||
1320 | mutex_exit(&os->os_lock); | |
1321 | ||
1322 | /* | |
1323 | * The dnode maintains a hold on its containing dbuf as | |
1324 | * long as there are holds on it. Each instantiated child | |
572e2857 | 1325 | * dbuf maintains a hold on the dnode. When the last child |
34dc7c2f BB |
1326 | * drops its hold, the dnode will drop its hold on the |
1327 | * containing dbuf. We add a "dirty hold" here so that the | |
1328 | * dnode will hang around after we finish processing its | |
1329 | * children. | |
1330 | */ | |
1331 | VERIFY(dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg)); | |
1332 | ||
1333 | (void) dbuf_dirty(dn->dn_dbuf, tx); | |
1334 | ||
1335 | dsl_dataset_dirty(os->os_dsl_dataset, tx); | |
1336 | } | |
1337 | ||
1338 | void | |
1339 | dnode_free(dnode_t *dn, dmu_tx_t *tx) | |
1340 | { | |
1341 | int txgoff = tx->tx_txg & TXG_MASK; | |
1342 | ||
1343 | dprintf("dn=%p txg=%llu\n", dn, tx->tx_txg); | |
1344 | ||
1345 | /* we should be the only holder... hopefully */ | |
1346 | /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */ | |
1347 | ||
1348 | mutex_enter(&dn->dn_mtx); | |
1349 | if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg) { | |
1350 | mutex_exit(&dn->dn_mtx); | |
1351 | return; | |
1352 | } | |
1353 | dn->dn_free_txg = tx->tx_txg; | |
1354 | mutex_exit(&dn->dn_mtx); | |
1355 | ||
1356 | /* | |
1357 | * If the dnode is already dirty, it needs to be moved from | |
1358 | * the dirty list to the free list. | |
1359 | */ | |
1360 | mutex_enter(&dn->dn_objset->os_lock); | |
1361 | if (list_link_active(&dn->dn_dirty_link[txgoff])) { | |
1362 | list_remove(&dn->dn_objset->os_dirty_dnodes[txgoff], dn); | |
1363 | list_insert_tail(&dn->dn_objset->os_free_dnodes[txgoff], dn); | |
1364 | mutex_exit(&dn->dn_objset->os_lock); | |
1365 | } else { | |
1366 | mutex_exit(&dn->dn_objset->os_lock); | |
1367 | dnode_setdirty(dn, tx); | |
1368 | } | |
1369 | } | |
1370 | ||
1371 | /* | |
1372 | * Try to change the block size for the indicated dnode. This can only | |
1373 | * succeed if there are no blocks allocated or dirty beyond first block | |
1374 | */ | |
1375 | int | |
1376 | dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) | |
1377 | { | |
8951cb8d | 1378 | dmu_buf_impl_t *db; |
b128c09f | 1379 | int err; |
34dc7c2f | 1380 | |
f1512ee6 | 1381 | ASSERT3U(size, <=, spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); |
34dc7c2f BB |
1382 | if (size == 0) |
1383 | size = SPA_MINBLOCKSIZE; | |
34dc7c2f BB |
1384 | else |
1385 | size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); | |
1386 | ||
1387 | if (ibs == dn->dn_indblkshift) | |
1388 | ibs = 0; | |
1389 | ||
1390 | if (size >> SPA_MINBLOCKSHIFT == dn->dn_datablkszsec && ibs == 0) | |
1391 | return (0); | |
1392 | ||
1393 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1394 | ||
1395 | /* Check for any allocated blocks beyond the first */ | |
93cf2076 | 1396 | if (dn->dn_maxblkid != 0) |
34dc7c2f BB |
1397 | goto fail; |
1398 | ||
1399 | mutex_enter(&dn->dn_dbufs_mtx); | |
8951cb8d AR |
1400 | for (db = avl_first(&dn->dn_dbufs); db != NULL; |
1401 | db = AVL_NEXT(&dn->dn_dbufs, db)) { | |
428870ff BB |
1402 | if (db->db_blkid != 0 && db->db_blkid != DMU_BONUS_BLKID && |
1403 | db->db_blkid != DMU_SPILL_BLKID) { | |
34dc7c2f BB |
1404 | mutex_exit(&dn->dn_dbufs_mtx); |
1405 | goto fail; | |
1406 | } | |
1407 | } | |
1408 | mutex_exit(&dn->dn_dbufs_mtx); | |
1409 | ||
1410 | if (ibs && dn->dn_nlevels != 1) | |
1411 | goto fail; | |
1412 | ||
b128c09f BB |
1413 | /* resize the old block */ |
1414 | err = dbuf_hold_impl(dn, 0, 0, TRUE, FTAG, &db); | |
1415 | if (err == 0) | |
34dc7c2f | 1416 | dbuf_new_size(db, size, tx); |
b128c09f BB |
1417 | else if (err != ENOENT) |
1418 | goto fail; | |
34dc7c2f BB |
1419 | |
1420 | dnode_setdblksz(dn, size); | |
1421 | dnode_setdirty(dn, tx); | |
1422 | dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size; | |
1423 | if (ibs) { | |
1424 | dn->dn_indblkshift = ibs; | |
1425 | dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; | |
1426 | } | |
b128c09f | 1427 | /* rele after we have fixed the blocksize in the dnode */ |
34dc7c2f BB |
1428 | if (db) |
1429 | dbuf_rele(db, FTAG); | |
1430 | ||
1431 | rw_exit(&dn->dn_struct_rwlock); | |
1432 | return (0); | |
1433 | ||
1434 | fail: | |
1435 | rw_exit(&dn->dn_struct_rwlock); | |
2e528b49 | 1436 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
1437 | } |
1438 | ||
b128c09f | 1439 | /* read-holding callers must not rely on the lock being continuously held */ |
34dc7c2f | 1440 | void |
b128c09f | 1441 | dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t have_read) |
34dc7c2f BB |
1442 | { |
1443 | uint64_t txgoff = tx->tx_txg & TXG_MASK; | |
34dc7c2f BB |
1444 | int epbs, new_nlevels; |
1445 | uint64_t sz; | |
1446 | ||
428870ff | 1447 | ASSERT(blkid != DMU_BONUS_BLKID); |
34dc7c2f | 1448 | |
b128c09f BB |
1449 | ASSERT(have_read ? |
1450 | RW_READ_HELD(&dn->dn_struct_rwlock) : | |
1451 | RW_WRITE_HELD(&dn->dn_struct_rwlock)); | |
1452 | ||
1453 | /* | |
1454 | * if we have a read-lock, check to see if we need to do any work | |
1455 | * before upgrading to a write-lock. | |
1456 | */ | |
1457 | if (have_read) { | |
1458 | if (blkid <= dn->dn_maxblkid) | |
1459 | return; | |
1460 | ||
1461 | if (!rw_tryupgrade(&dn->dn_struct_rwlock)) { | |
1462 | rw_exit(&dn->dn_struct_rwlock); | |
1463 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1464 | } | |
34dc7c2f BB |
1465 | } |
1466 | ||
1467 | if (blkid <= dn->dn_maxblkid) | |
1468 | goto out; | |
1469 | ||
1470 | dn->dn_maxblkid = blkid; | |
1471 | ||
1472 | /* | |
1473 | * Compute the number of levels necessary to support the new maxblkid. | |
1474 | */ | |
1475 | new_nlevels = 1; | |
1476 | epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
1477 | for (sz = dn->dn_nblkptr; | |
1478 | sz <= blkid && sz >= dn->dn_nblkptr; sz <<= epbs) | |
1479 | new_nlevels++; | |
1480 | ||
1481 | if (new_nlevels > dn->dn_nlevels) { | |
1482 | int old_nlevels = dn->dn_nlevels; | |
1483 | dmu_buf_impl_t *db; | |
1484 | list_t *list; | |
1485 | dbuf_dirty_record_t *new, *dr, *dr_next; | |
1486 | ||
1487 | dn->dn_nlevels = new_nlevels; | |
1488 | ||
1489 | ASSERT3U(new_nlevels, >, dn->dn_next_nlevels[txgoff]); | |
1490 | dn->dn_next_nlevels[txgoff] = new_nlevels; | |
1491 | ||
1492 | /* dirty the left indirects */ | |
1493 | db = dbuf_hold_level(dn, old_nlevels, 0, FTAG); | |
428870ff | 1494 | ASSERT(db != NULL); |
34dc7c2f BB |
1495 | new = dbuf_dirty(db, tx); |
1496 | dbuf_rele(db, FTAG); | |
1497 | ||
1498 | /* transfer the dirty records to the new indirect */ | |
1499 | mutex_enter(&dn->dn_mtx); | |
1500 | mutex_enter(&new->dt.di.dr_mtx); | |
1501 | list = &dn->dn_dirty_records[txgoff]; | |
1502 | for (dr = list_head(list); dr; dr = dr_next) { | |
1503 | dr_next = list_next(&dn->dn_dirty_records[txgoff], dr); | |
1504 | if (dr->dr_dbuf->db_level != new_nlevels-1 && | |
428870ff BB |
1505 | dr->dr_dbuf->db_blkid != DMU_BONUS_BLKID && |
1506 | dr->dr_dbuf->db_blkid != DMU_SPILL_BLKID) { | |
34dc7c2f BB |
1507 | ASSERT(dr->dr_dbuf->db_level == old_nlevels-1); |
1508 | list_remove(&dn->dn_dirty_records[txgoff], dr); | |
1509 | list_insert_tail(&new->dt.di.dr_children, dr); | |
1510 | dr->dr_parent = new; | |
1511 | } | |
1512 | } | |
1513 | mutex_exit(&new->dt.di.dr_mtx); | |
1514 | mutex_exit(&dn->dn_mtx); | |
1515 | } | |
1516 | ||
1517 | out: | |
b128c09f BB |
1518 | if (have_read) |
1519 | rw_downgrade(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
1520 | } |
1521 | ||
34dc7c2f BB |
1522 | void |
1523 | dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx) | |
1524 | { | |
1525 | dmu_buf_impl_t *db; | |
1526 | uint64_t blkoff, blkid, nblks; | |
b128c09f | 1527 | int blksz, blkshift, head, tail; |
34dc7c2f | 1528 | int trunc = FALSE; |
b128c09f | 1529 | int epbs; |
34dc7c2f BB |
1530 | |
1531 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1532 | blksz = dn->dn_datablksz; | |
b128c09f BB |
1533 | blkshift = dn->dn_datablkshift; |
1534 | epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
34dc7c2f | 1535 | |
b663a23d | 1536 | if (len == DMU_OBJECT_END) { |
34dc7c2f BB |
1537 | len = UINT64_MAX - off; |
1538 | trunc = TRUE; | |
1539 | } | |
1540 | ||
1541 | /* | |
1542 | * First, block align the region to free: | |
1543 | */ | |
1544 | if (ISP2(blksz)) { | |
1545 | head = P2NPHASE(off, blksz); | |
1546 | blkoff = P2PHASE(off, blksz); | |
b128c09f BB |
1547 | if ((off >> blkshift) > dn->dn_maxblkid) |
1548 | goto out; | |
34dc7c2f BB |
1549 | } else { |
1550 | ASSERT(dn->dn_maxblkid == 0); | |
1551 | if (off == 0 && len >= blksz) { | |
b0bc7a84 MG |
1552 | /* |
1553 | * Freeing the whole block; fast-track this request. | |
1554 | * Note that we won't dirty any indirect blocks, | |
1555 | * which is fine because we will be freeing the entire | |
1556 | * file and thus all indirect blocks will be freed | |
1557 | * by free_children(). | |
1558 | */ | |
b128c09f BB |
1559 | blkid = 0; |
1560 | nblks = 1; | |
1561 | goto done; | |
1562 | } else if (off >= blksz) { | |
1563 | /* Freeing past end-of-data */ | |
1564 | goto out; | |
34dc7c2f BB |
1565 | } else { |
1566 | /* Freeing part of the block. */ | |
1567 | head = blksz - off; | |
1568 | ASSERT3U(head, >, 0); | |
1569 | } | |
1570 | blkoff = off; | |
1571 | } | |
1572 | /* zero out any partial block data at the start of the range */ | |
1573 | if (head) { | |
1574 | ASSERT3U(blkoff + head, ==, blksz); | |
1575 | if (len < head) | |
1576 | head = len; | |
1577 | if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off), TRUE, | |
1578 | FTAG, &db) == 0) { | |
1579 | caddr_t data; | |
1580 | ||
1581 | /* don't dirty if it isn't on disk and isn't dirty */ | |
1582 | if (db->db_last_dirty || | |
1583 | (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { | |
1584 | rw_exit(&dn->dn_struct_rwlock); | |
b0bc7a84 | 1585 | dmu_buf_will_dirty(&db->db, tx); |
34dc7c2f BB |
1586 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); |
1587 | data = db->db.db_data; | |
1588 | bzero(data + blkoff, head); | |
1589 | } | |
1590 | dbuf_rele(db, FTAG); | |
1591 | } | |
1592 | off += head; | |
1593 | len -= head; | |
1594 | } | |
1595 | ||
1596 | /* If the range was less than one block, we're done */ | |
b128c09f | 1597 | if (len == 0) |
34dc7c2f BB |
1598 | goto out; |
1599 | ||
b128c09f BB |
1600 | /* If the remaining range is past end of file, we're done */ |
1601 | if ((off >> blkshift) > dn->dn_maxblkid) | |
1602 | goto out; | |
34dc7c2f | 1603 | |
b128c09f BB |
1604 | ASSERT(ISP2(blksz)); |
1605 | if (trunc) | |
1606 | tail = 0; | |
1607 | else | |
1608 | tail = P2PHASE(len, blksz); | |
1609 | ||
c99c9001 | 1610 | ASSERT0(P2PHASE(off, blksz)); |
b128c09f BB |
1611 | /* zero out any partial block data at the end of the range */ |
1612 | if (tail) { | |
1613 | if (len < tail) | |
1614 | tail = len; | |
1615 | if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off+len), | |
1616 | TRUE, FTAG, &db) == 0) { | |
1617 | /* don't dirty if not on disk and not dirty */ | |
1618 | if (db->db_last_dirty || | |
1619 | (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { | |
1620 | rw_exit(&dn->dn_struct_rwlock); | |
b0bc7a84 | 1621 | dmu_buf_will_dirty(&db->db, tx); |
b128c09f BB |
1622 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); |
1623 | bzero(db->db.db_data, tail); | |
34dc7c2f | 1624 | } |
b128c09f | 1625 | dbuf_rele(db, FTAG); |
34dc7c2f | 1626 | } |
b128c09f BB |
1627 | len -= tail; |
1628 | } | |
34dc7c2f | 1629 | |
b128c09f BB |
1630 | /* If the range did not include a full block, we are done */ |
1631 | if (len == 0) | |
1632 | goto out; | |
1633 | ||
1634 | ASSERT(IS_P2ALIGNED(off, blksz)); | |
1635 | ASSERT(trunc || IS_P2ALIGNED(len, blksz)); | |
1636 | blkid = off >> blkshift; | |
1637 | nblks = len >> blkshift; | |
1638 | if (trunc) | |
1639 | nblks += 1; | |
1640 | ||
1641 | /* | |
b0bc7a84 MG |
1642 | * Dirty the first and last indirect blocks, as they (and/or their |
1643 | * parents) will need to be written out if they were only | |
1644 | * partially freed. Interior indirect blocks will be themselves freed, | |
1645 | * by free_children(), so they need not be dirtied. Note that these | |
1646 | * interior blocks have already been prefetched by dmu_tx_hold_free(). | |
b128c09f BB |
1647 | */ |
1648 | if (dn->dn_nlevels > 1) { | |
b0bc7a84 | 1649 | uint64_t first, last; |
b128c09f BB |
1650 | |
1651 | first = blkid >> epbs; | |
c65aa5b2 | 1652 | if ((db = dbuf_hold_level(dn, 1, first, FTAG))) { |
b0bc7a84 | 1653 | dmu_buf_will_dirty(&db->db, tx); |
34dc7c2f BB |
1654 | dbuf_rele(db, FTAG); |
1655 | } | |
b128c09f BB |
1656 | if (trunc) |
1657 | last = dn->dn_maxblkid >> epbs; | |
1658 | else | |
1659 | last = (blkid + nblks - 1) >> epbs; | |
1660 | if (last > first && (db = dbuf_hold_level(dn, 1, last, FTAG))) { | |
b0bc7a84 | 1661 | dmu_buf_will_dirty(&db->db, tx); |
34dc7c2f BB |
1662 | dbuf_rele(db, FTAG); |
1663 | } | |
34dc7c2f | 1664 | } |
b0bc7a84 | 1665 | |
b128c09f BB |
1666 | done: |
1667 | /* | |
1668 | * Add this range to the dnode range list. | |
1669 | * We will finish up this free operation in the syncing phase. | |
1670 | */ | |
34dc7c2f | 1671 | mutex_enter(&dn->dn_mtx); |
34dc7c2f | 1672 | { |
9bd274dd MA |
1673 | int txgoff = tx->tx_txg & TXG_MASK; |
1674 | if (dn->dn_free_ranges[txgoff] == NULL) { | |
1675 | dn->dn_free_ranges[txgoff] = | |
1676 | range_tree_create(NULL, NULL, &dn->dn_mtx); | |
1677 | } | |
1678 | range_tree_clear(dn->dn_free_ranges[txgoff], blkid, nblks); | |
1679 | range_tree_add(dn->dn_free_ranges[txgoff], blkid, nblks); | |
34dc7c2f | 1680 | } |
9bd274dd MA |
1681 | dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", |
1682 | blkid, nblks, tx->tx_txg); | |
34dc7c2f BB |
1683 | mutex_exit(&dn->dn_mtx); |
1684 | ||
b128c09f | 1685 | dbuf_free_range(dn, blkid, blkid + nblks - 1, tx); |
34dc7c2f BB |
1686 | dnode_setdirty(dn, tx); |
1687 | out: | |
b128c09f | 1688 | |
34dc7c2f BB |
1689 | rw_exit(&dn->dn_struct_rwlock); |
1690 | } | |
1691 | ||
428870ff BB |
1692 | static boolean_t |
1693 | dnode_spill_freed(dnode_t *dn) | |
1694 | { | |
1695 | int i; | |
1696 | ||
1697 | mutex_enter(&dn->dn_mtx); | |
1698 | for (i = 0; i < TXG_SIZE; i++) { | |
1699 | if (dn->dn_rm_spillblk[i] == DN_KILL_SPILLBLK) | |
1700 | break; | |
1701 | } | |
1702 | mutex_exit(&dn->dn_mtx); | |
1703 | return (i < TXG_SIZE); | |
1704 | } | |
1705 | ||
34dc7c2f BB |
1706 | /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */ |
1707 | uint64_t | |
1708 | dnode_block_freed(dnode_t *dn, uint64_t blkid) | |
1709 | { | |
34dc7c2f BB |
1710 | void *dp = spa_get_dsl(dn->dn_objset->os_spa); |
1711 | int i; | |
1712 | ||
428870ff | 1713 | if (blkid == DMU_BONUS_BLKID) |
34dc7c2f BB |
1714 | return (FALSE); |
1715 | ||
1716 | /* | |
1717 | * If we're in the process of opening the pool, dp will not be | |
1718 | * set yet, but there shouldn't be anything dirty. | |
1719 | */ | |
1720 | if (dp == NULL) | |
1721 | return (FALSE); | |
1722 | ||
1723 | if (dn->dn_free_txg) | |
1724 | return (TRUE); | |
1725 | ||
428870ff BB |
1726 | if (blkid == DMU_SPILL_BLKID) |
1727 | return (dnode_spill_freed(dn)); | |
1728 | ||
34dc7c2f BB |
1729 | mutex_enter(&dn->dn_mtx); |
1730 | for (i = 0; i < TXG_SIZE; i++) { | |
9bd274dd MA |
1731 | if (dn->dn_free_ranges[i] != NULL && |
1732 | range_tree_contains(dn->dn_free_ranges[i], blkid, 1)) | |
34dc7c2f BB |
1733 | break; |
1734 | } | |
1735 | mutex_exit(&dn->dn_mtx); | |
1736 | return (i < TXG_SIZE); | |
1737 | } | |
1738 | ||
1739 | /* call from syncing context when we actually write/free space for this dnode */ | |
1740 | void | |
1741 | dnode_diduse_space(dnode_t *dn, int64_t delta) | |
1742 | { | |
1743 | uint64_t space; | |
1744 | dprintf_dnode(dn, "dn=%p dnp=%p used=%llu delta=%lld\n", | |
1745 | dn, dn->dn_phys, | |
1746 | (u_longlong_t)dn->dn_phys->dn_used, | |
1747 | (longlong_t)delta); | |
1748 | ||
1749 | mutex_enter(&dn->dn_mtx); | |
1750 | space = DN_USED_BYTES(dn->dn_phys); | |
1751 | if (delta > 0) { | |
1752 | ASSERT3U(space + delta, >=, space); /* no overflow */ | |
1753 | } else { | |
1754 | ASSERT3U(space, >=, -delta); /* no underflow */ | |
1755 | } | |
1756 | space += delta; | |
1757 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_DNODE_BYTES) { | |
1758 | ASSERT((dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) == 0); | |
c99c9001 | 1759 | ASSERT0(P2PHASE(space, 1<<DEV_BSHIFT)); |
34dc7c2f BB |
1760 | dn->dn_phys->dn_used = space >> DEV_BSHIFT; |
1761 | } else { | |
1762 | dn->dn_phys->dn_used = space; | |
1763 | dn->dn_phys->dn_flags |= DNODE_FLAG_USED_BYTES; | |
1764 | } | |
1765 | mutex_exit(&dn->dn_mtx); | |
1766 | } | |
1767 | ||
1768 | /* | |
e8b96c60 MA |
1769 | * Call when we think we're going to write/free space in open context to track |
1770 | * the amount of memory in use by the currently open txg. | |
34dc7c2f BB |
1771 | */ |
1772 | void | |
1773 | dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx) | |
1774 | { | |
428870ff | 1775 | objset_t *os = dn->dn_objset; |
34dc7c2f | 1776 | dsl_dataset_t *ds = os->os_dsl_dataset; |
e8b96c60 | 1777 | int64_t aspace = spa_get_asize(os->os_spa, space); |
34dc7c2f | 1778 | |
e8b96c60 MA |
1779 | if (ds != NULL) { |
1780 | dsl_dir_willuse_space(ds->ds_dir, aspace, tx); | |
1781 | dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); | |
1782 | } | |
34dc7c2f | 1783 | |
e8b96c60 | 1784 | dmu_tx_willuse_space(tx, aspace); |
34dc7c2f BB |
1785 | } |
1786 | ||
45d1cae3 | 1787 | /* |
d3cc8b15 WA |
1788 | * Scans a block at the indicated "level" looking for a hole or data, |
1789 | * depending on 'flags'. | |
1790 | * | |
1791 | * If level > 0, then we are scanning an indirect block looking at its | |
1792 | * pointers. If level == 0, then we are looking at a block of dnodes. | |
1793 | * | |
1794 | * If we don't find what we are looking for in the block, we return ESRCH. | |
1795 | * Otherwise, return with *offset pointing to the beginning (if searching | |
1796 | * forwards) or end (if searching backwards) of the range covered by the | |
1797 | * block pointer we matched on (or dnode). | |
45d1cae3 BB |
1798 | * |
1799 | * The basic search algorithm used below by dnode_next_offset() is to | |
1800 | * use this function to search up the block tree (widen the search) until | |
1801 | * we find something (i.e., we don't return ESRCH) and then search back | |
1802 | * down the tree (narrow the search) until we reach our original search | |
1803 | * level. | |
1804 | */ | |
34dc7c2f | 1805 | static int |
b128c09f | 1806 | dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset, |
34dc7c2f BB |
1807 | int lvl, uint64_t blkfill, uint64_t txg) |
1808 | { | |
1809 | dmu_buf_impl_t *db = NULL; | |
1810 | void *data = NULL; | |
1811 | uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
1812 | uint64_t epb = 1ULL << epbs; | |
1813 | uint64_t minfill, maxfill; | |
b128c09f BB |
1814 | boolean_t hole; |
1815 | int i, inc, error, span; | |
34dc7c2f BB |
1816 | |
1817 | dprintf("probing object %llu offset %llx level %d of %u\n", | |
1818 | dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels); | |
1819 | ||
9babb374 | 1820 | hole = ((flags & DNODE_FIND_HOLE) != 0); |
b128c09f BB |
1821 | inc = (flags & DNODE_FIND_BACKWARDS) ? -1 : 1; |
1822 | ASSERT(txg == 0 || !hole); | |
1823 | ||
34dc7c2f BB |
1824 | if (lvl == dn->dn_phys->dn_nlevels) { |
1825 | error = 0; | |
1826 | epb = dn->dn_phys->dn_nblkptr; | |
1827 | data = dn->dn_phys->dn_blkptr; | |
1828 | } else { | |
1829 | uint64_t blkid = dbuf_whichblock(dn, *offset) >> (epbs * lvl); | |
1830 | error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FTAG, &db); | |
1831 | if (error) { | |
b128c09f BB |
1832 | if (error != ENOENT) |
1833 | return (error); | |
1834 | if (hole) | |
1835 | return (0); | |
1836 | /* | |
1837 | * This can only happen when we are searching up | |
1838 | * the block tree for data. We don't really need to | |
1839 | * adjust the offset, as we will just end up looking | |
1840 | * at the pointer to this block in its parent, and its | |
1841 | * going to be unallocated, so we will skip over it. | |
1842 | */ | |
2e528b49 | 1843 | return (SET_ERROR(ESRCH)); |
34dc7c2f BB |
1844 | } |
1845 | error = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_HAVESTRUCT); | |
1846 | if (error) { | |
1847 | dbuf_rele(db, FTAG); | |
1848 | return (error); | |
1849 | } | |
1850 | data = db->db.db_data; | |
1851 | } | |
1852 | ||
b0bc7a84 MG |
1853 | |
1854 | if (db != NULL && txg != 0 && (db->db_blkptr == NULL || | |
1855 | db->db_blkptr->blk_birth <= txg || | |
1856 | BP_IS_HOLE(db->db_blkptr))) { | |
b128c09f BB |
1857 | /* |
1858 | * This can only happen when we are searching up the tree | |
1859 | * and these conditions mean that we need to keep climbing. | |
1860 | */ | |
2e528b49 | 1861 | error = SET_ERROR(ESRCH); |
34dc7c2f BB |
1862 | } else if (lvl == 0) { |
1863 | dnode_phys_t *dnp = data; | |
1864 | span = DNODE_SHIFT; | |
1865 | ASSERT(dn->dn_type == DMU_OT_DNODE); | |
1866 | ||
b128c09f BB |
1867 | for (i = (*offset >> span) & (blkfill - 1); |
1868 | i >= 0 && i < blkfill; i += inc) { | |
9babb374 | 1869 | if ((dnp[i].dn_type == DMU_OT_NONE) == hole) |
34dc7c2f | 1870 | break; |
b128c09f | 1871 | *offset += (1ULL << span) * inc; |
34dc7c2f | 1872 | } |
b128c09f | 1873 | if (i < 0 || i == blkfill) |
2e528b49 | 1874 | error = SET_ERROR(ESRCH); |
34dc7c2f BB |
1875 | } else { |
1876 | blkptr_t *bp = data; | |
45d1cae3 | 1877 | uint64_t start = *offset; |
34dc7c2f BB |
1878 | span = (lvl - 1) * epbs + dn->dn_datablkshift; |
1879 | minfill = 0; | |
1880 | maxfill = blkfill << ((lvl - 1) * epbs); | |
1881 | ||
1882 | if (hole) | |
1883 | maxfill--; | |
1884 | else | |
1885 | minfill++; | |
1886 | ||
45d1cae3 BB |
1887 | *offset = *offset >> span; |
1888 | for (i = BF64_GET(*offset, 0, epbs); | |
b128c09f | 1889 | i >= 0 && i < epb; i += inc) { |
9b67f605 MA |
1890 | if (BP_GET_FILL(&bp[i]) >= minfill && |
1891 | BP_GET_FILL(&bp[i]) <= maxfill && | |
b128c09f | 1892 | (hole || bp[i].blk_birth > txg)) |
34dc7c2f | 1893 | break; |
45d1cae3 BB |
1894 | if (inc > 0 || *offset > 0) |
1895 | *offset += inc; | |
1896 | } | |
1897 | *offset = *offset << span; | |
1898 | if (inc < 0) { | |
1899 | /* traversing backwards; position offset at the end */ | |
1900 | ASSERT3U(*offset, <=, start); | |
1901 | *offset = MIN(*offset + (1ULL << span) - 1, start); | |
1902 | } else if (*offset < start) { | |
1903 | *offset = start; | |
34dc7c2f | 1904 | } |
45d1cae3 | 1905 | if (i < 0 || i >= epb) |
2e528b49 | 1906 | error = SET_ERROR(ESRCH); |
34dc7c2f BB |
1907 | } |
1908 | ||
1909 | if (db) | |
1910 | dbuf_rele(db, FTAG); | |
1911 | ||
1912 | return (error); | |
1913 | } | |
1914 | ||
1915 | /* | |
1916 | * Find the next hole, data, or sparse region at or after *offset. | |
1917 | * The value 'blkfill' tells us how many items we expect to find | |
1918 | * in an L0 data block; this value is 1 for normal objects, | |
1919 | * DNODES_PER_BLOCK for the meta dnode, and some fraction of | |
1920 | * DNODES_PER_BLOCK when searching for sparse regions thereof. | |
1921 | * | |
1922 | * Examples: | |
1923 | * | |
b128c09f BB |
1924 | * dnode_next_offset(dn, flags, offset, 1, 1, 0); |
1925 | * Finds the next/previous hole/data in a file. | |
34dc7c2f BB |
1926 | * Used in dmu_offset_next(). |
1927 | * | |
b128c09f | 1928 | * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg); |
34dc7c2f BB |
1929 | * Finds the next free/allocated dnode an objset's meta-dnode. |
1930 | * Only finds objects that have new contents since txg (ie. | |
1931 | * bonus buffer changes and content removal are ignored). | |
1932 | * Used in dmu_object_next(). | |
1933 | * | |
b128c09f | 1934 | * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0); |
34dc7c2f BB |
1935 | * Finds the next L2 meta-dnode bp that's at most 1/4 full. |
1936 | * Used in dmu_object_alloc(). | |
1937 | */ | |
1938 | int | |
b128c09f | 1939 | dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset, |
34dc7c2f BB |
1940 | int minlvl, uint64_t blkfill, uint64_t txg) |
1941 | { | |
b128c09f | 1942 | uint64_t initial_offset = *offset; |
34dc7c2f BB |
1943 | int lvl, maxlvl; |
1944 | int error = 0; | |
34dc7c2f | 1945 | |
b128c09f BB |
1946 | if (!(flags & DNODE_FIND_HAVELOCK)) |
1947 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
34dc7c2f BB |
1948 | |
1949 | if (dn->dn_phys->dn_nlevels == 0) { | |
2e528b49 | 1950 | error = SET_ERROR(ESRCH); |
b128c09f | 1951 | goto out; |
34dc7c2f BB |
1952 | } |
1953 | ||
1954 | if (dn->dn_datablkshift == 0) { | |
1955 | if (*offset < dn->dn_datablksz) { | |
b128c09f | 1956 | if (flags & DNODE_FIND_HOLE) |
34dc7c2f BB |
1957 | *offset = dn->dn_datablksz; |
1958 | } else { | |
2e528b49 | 1959 | error = SET_ERROR(ESRCH); |
34dc7c2f | 1960 | } |
b128c09f | 1961 | goto out; |
34dc7c2f BB |
1962 | } |
1963 | ||
1964 | maxlvl = dn->dn_phys->dn_nlevels; | |
1965 | ||
1966 | for (lvl = minlvl; lvl <= maxlvl; lvl++) { | |
1967 | error = dnode_next_offset_level(dn, | |
b128c09f | 1968 | flags, offset, lvl, blkfill, txg); |
34dc7c2f BB |
1969 | if (error != ESRCH) |
1970 | break; | |
1971 | } | |
1972 | ||
b128c09f | 1973 | while (error == 0 && --lvl >= minlvl) { |
34dc7c2f | 1974 | error = dnode_next_offset_level(dn, |
b128c09f | 1975 | flags, offset, lvl, blkfill, txg); |
34dc7c2f BB |
1976 | } |
1977 | ||
d97aa48f MA |
1978 | /* |
1979 | * There's always a "virtual hole" at the end of the object, even | |
1980 | * if all BP's which physically exist are non-holes. | |
1981 | */ | |
1982 | if ((flags & DNODE_FIND_HOLE) && error == ESRCH && txg == 0 && | |
1983 | minlvl == 1 && blkfill == 1 && !(flags & DNODE_FIND_BACKWARDS)) { | |
1984 | error = 0; | |
1985 | } | |
1986 | ||
b128c09f BB |
1987 | if (error == 0 && (flags & DNODE_FIND_BACKWARDS ? |
1988 | initial_offset < *offset : initial_offset > *offset)) | |
2e528b49 | 1989 | error = SET_ERROR(ESRCH); |
b128c09f BB |
1990 | out: |
1991 | if (!(flags & DNODE_FIND_HAVELOCK)) | |
1992 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
1993 | |
1994 | return (error); | |
1995 | } |