]>
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. |
34dc7c2f BB |
24 | */ |
25 | ||
34dc7c2f BB |
26 | #include <sys/zfs_context.h> |
27 | #include <sys/dbuf.h> | |
28 | #include <sys/dnode.h> | |
29 | #include <sys/dmu.h> | |
30 | #include <sys/dmu_impl.h> | |
31 | #include <sys/dmu_tx.h> | |
32 | #include <sys/dmu_objset.h> | |
33 | #include <sys/dsl_dir.h> | |
34 | #include <sys/dsl_dataset.h> | |
35 | #include <sys/spa.h> | |
36 | #include <sys/zio.h> | |
37 | #include <sys/dmu_zfetch.h> | |
9bd274dd | 38 | #include <sys/range_tree.h> |
49ee64e5 | 39 | #include <sys/trace_dnode.h> |
34dc7c2f BB |
40 | |
41 | static kmem_cache_t *dnode_cache; | |
572e2857 BB |
42 | /* |
43 | * Define DNODE_STATS to turn on statistic gathering. By default, it is only | |
44 | * turned on when DEBUG is also defined. | |
45 | */ | |
46 | #ifdef DEBUG | |
47 | #define DNODE_STATS | |
48 | #endif /* DEBUG */ | |
49 | ||
50 | #ifdef DNODE_STATS | |
51 | #define DNODE_STAT_ADD(stat) ((stat)++) | |
52 | #else | |
53 | #define DNODE_STAT_ADD(stat) /* nothing */ | |
54 | #endif /* DNODE_STATS */ | |
34dc7c2f | 55 | |
1fde1e37 | 56 | ASSERTV(static dnode_phys_t dnode_phys_zero); |
34dc7c2f BB |
57 | |
58 | int zfs_default_bs = SPA_MINBLOCKSHIFT; | |
59 | int zfs_default_ibs = DN_MAX_INDBLKSHIFT; | |
60 | ||
5ac1241a | 61 | #ifdef _KERNEL |
572e2857 | 62 | static kmem_cbrc_t dnode_move(void *, void *, size_t, void *); |
5ac1241a | 63 | #endif /* _KERNEL */ |
572e2857 | 64 | |
34dc7c2f BB |
65 | /* ARGSUSED */ |
66 | static int | |
67 | dnode_cons(void *arg, void *unused, int kmflag) | |
68 | { | |
34dc7c2f | 69 | dnode_t *dn = arg; |
572e2857 | 70 | int i; |
34dc7c2f BB |
71 | |
72 | rw_init(&dn->dn_struct_rwlock, NULL, RW_DEFAULT, NULL); | |
73 | mutex_init(&dn->dn_mtx, NULL, MUTEX_DEFAULT, NULL); | |
74 | mutex_init(&dn->dn_dbufs_mtx, NULL, MUTEX_DEFAULT, NULL); | |
fb5f0bc8 BB |
75 | cv_init(&dn->dn_notxholds, NULL, CV_DEFAULT, NULL); |
76 | ||
13fe0198 MA |
77 | /* |
78 | * Every dbuf has a reference, and dropping a tracked reference is | |
79 | * O(number of references), so don't track dn_holds. | |
80 | */ | |
81 | refcount_create_untracked(&dn->dn_holds); | |
34dc7c2f | 82 | refcount_create(&dn->dn_tx_holds); |
572e2857 BB |
83 | list_link_init(&dn->dn_link); |
84 | ||
85 | bzero(&dn->dn_next_nblkptr[0], sizeof (dn->dn_next_nblkptr)); | |
86 | bzero(&dn->dn_next_nlevels[0], sizeof (dn->dn_next_nlevels)); | |
87 | bzero(&dn->dn_next_indblkshift[0], sizeof (dn->dn_next_indblkshift)); | |
88 | bzero(&dn->dn_next_bonustype[0], sizeof (dn->dn_next_bonustype)); | |
89 | bzero(&dn->dn_rm_spillblk[0], sizeof (dn->dn_rm_spillblk)); | |
90 | bzero(&dn->dn_next_bonuslen[0], sizeof (dn->dn_next_bonuslen)); | |
91 | bzero(&dn->dn_next_blksz[0], sizeof (dn->dn_next_blksz)); | |
34dc7c2f BB |
92 | |
93 | for (i = 0; i < TXG_SIZE; i++) { | |
572e2857 | 94 | list_link_init(&dn->dn_dirty_link[i]); |
9bd274dd | 95 | dn->dn_free_ranges[i] = NULL; |
34dc7c2f BB |
96 | list_create(&dn->dn_dirty_records[i], |
97 | sizeof (dbuf_dirty_record_t), | |
98 | offsetof(dbuf_dirty_record_t, dr_dirty_node)); | |
99 | } | |
100 | ||
572e2857 BB |
101 | dn->dn_allocated_txg = 0; |
102 | dn->dn_free_txg = 0; | |
103 | dn->dn_assigned_txg = 0; | |
104 | dn->dn_dirtyctx = 0; | |
105 | dn->dn_dirtyctx_firstset = NULL; | |
106 | dn->dn_bonus = NULL; | |
107 | dn->dn_have_spill = B_FALSE; | |
108 | dn->dn_zio = NULL; | |
109 | dn->dn_oldused = 0; | |
110 | dn->dn_oldflags = 0; | |
111 | dn->dn_olduid = 0; | |
112 | dn->dn_oldgid = 0; | |
113 | dn->dn_newuid = 0; | |
114 | dn->dn_newgid = 0; | |
115 | dn->dn_id_flags = 0; | |
116 | ||
117 | dn->dn_dbufs_count = 0; | |
b663a23d | 118 | dn->dn_unlisted_l0_blkid = 0; |
34dc7c2f BB |
119 | list_create(&dn->dn_dbufs, sizeof (dmu_buf_impl_t), |
120 | offsetof(dmu_buf_impl_t, db_link)); | |
121 | ||
572e2857 | 122 | dn->dn_moved = 0; |
34dc7c2f BB |
123 | return (0); |
124 | } | |
125 | ||
126 | /* ARGSUSED */ | |
127 | static void | |
128 | dnode_dest(void *arg, void *unused) | |
129 | { | |
130 | int i; | |
131 | dnode_t *dn = arg; | |
132 | ||
133 | rw_destroy(&dn->dn_struct_rwlock); | |
134 | mutex_destroy(&dn->dn_mtx); | |
135 | mutex_destroy(&dn->dn_dbufs_mtx); | |
fb5f0bc8 | 136 | cv_destroy(&dn->dn_notxholds); |
34dc7c2f BB |
137 | refcount_destroy(&dn->dn_holds); |
138 | refcount_destroy(&dn->dn_tx_holds); | |
572e2857 | 139 | ASSERT(!list_link_active(&dn->dn_link)); |
34dc7c2f BB |
140 | |
141 | for (i = 0; i < TXG_SIZE; i++) { | |
572e2857 | 142 | ASSERT(!list_link_active(&dn->dn_dirty_link[i])); |
9bd274dd | 143 | ASSERT3P(dn->dn_free_ranges[i], ==, NULL); |
34dc7c2f | 144 | list_destroy(&dn->dn_dirty_records[i]); |
c99c9001 MS |
145 | ASSERT0(dn->dn_next_nblkptr[i]); |
146 | ASSERT0(dn->dn_next_nlevels[i]); | |
147 | ASSERT0(dn->dn_next_indblkshift[i]); | |
148 | ASSERT0(dn->dn_next_bonustype[i]); | |
149 | ASSERT0(dn->dn_rm_spillblk[i]); | |
150 | ASSERT0(dn->dn_next_bonuslen[i]); | |
151 | ASSERT0(dn->dn_next_blksz[i]); | |
34dc7c2f BB |
152 | } |
153 | ||
c99c9001 MS |
154 | ASSERT0(dn->dn_allocated_txg); |
155 | ASSERT0(dn->dn_free_txg); | |
156 | ASSERT0(dn->dn_assigned_txg); | |
157 | ASSERT0(dn->dn_dirtyctx); | |
572e2857 BB |
158 | ASSERT3P(dn->dn_dirtyctx_firstset, ==, NULL); |
159 | ASSERT3P(dn->dn_bonus, ==, NULL); | |
160 | ASSERT(!dn->dn_have_spill); | |
161 | ASSERT3P(dn->dn_zio, ==, NULL); | |
c99c9001 MS |
162 | ASSERT0(dn->dn_oldused); |
163 | ASSERT0(dn->dn_oldflags); | |
164 | ASSERT0(dn->dn_olduid); | |
165 | ASSERT0(dn->dn_oldgid); | |
166 | ASSERT0(dn->dn_newuid); | |
167 | ASSERT0(dn->dn_newgid); | |
168 | ASSERT0(dn->dn_id_flags); | |
169 | ||
170 | ASSERT0(dn->dn_dbufs_count); | |
b663a23d | 171 | ASSERT0(dn->dn_unlisted_l0_blkid); |
34dc7c2f BB |
172 | list_destroy(&dn->dn_dbufs); |
173 | } | |
174 | ||
175 | void | |
176 | dnode_init(void) | |
177 | { | |
572e2857 | 178 | ASSERT(dnode_cache == NULL); |
ae6ba3db | 179 | dnode_cache = kmem_cache_create("dnode_t", sizeof (dnode_t), |
6795a698 | 180 | 0, dnode_cons, dnode_dest, NULL, NULL, NULL, 0); |
572e2857 | 181 | kmem_cache_set_move(dnode_cache, dnode_move); |
34dc7c2f BB |
182 | } |
183 | ||
184 | void | |
185 | dnode_fini(void) | |
186 | { | |
187 | kmem_cache_destroy(dnode_cache); | |
572e2857 | 188 | dnode_cache = NULL; |
34dc7c2f BB |
189 | } |
190 | ||
191 | ||
192 | #ifdef ZFS_DEBUG | |
193 | void | |
194 | dnode_verify(dnode_t *dn) | |
195 | { | |
196 | int drop_struct_lock = FALSE; | |
197 | ||
198 | ASSERT(dn->dn_phys); | |
199 | ASSERT(dn->dn_objset); | |
572e2857 | 200 | ASSERT(dn->dn_handle->dnh_dnode == dn); |
34dc7c2f | 201 | |
9ae529ec | 202 | ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type)); |
34dc7c2f BB |
203 | |
204 | if (!(zfs_flags & ZFS_DEBUG_DNODE_VERIFY)) | |
205 | return; | |
206 | ||
207 | if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { | |
208 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
209 | drop_struct_lock = TRUE; | |
210 | } | |
211 | if (dn->dn_phys->dn_type != DMU_OT_NONE || dn->dn_allocated_txg != 0) { | |
212 | int i; | |
34dc7c2f BB |
213 | ASSERT3U(dn->dn_indblkshift, <=, SPA_MAXBLOCKSHIFT); |
214 | if (dn->dn_datablkshift) { | |
215 | ASSERT3U(dn->dn_datablkshift, >=, SPA_MINBLOCKSHIFT); | |
216 | ASSERT3U(dn->dn_datablkshift, <=, SPA_MAXBLOCKSHIFT); | |
217 | ASSERT3U(1<<dn->dn_datablkshift, ==, dn->dn_datablksz); | |
218 | } | |
219 | ASSERT3U(dn->dn_nlevels, <=, 30); | |
9ae529ec | 220 | ASSERT(DMU_OT_IS_VALID(dn->dn_type)); |
34dc7c2f BB |
221 | ASSERT3U(dn->dn_nblkptr, >=, 1); |
222 | ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); | |
223 | ASSERT3U(dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); | |
224 | ASSERT3U(dn->dn_datablksz, ==, | |
225 | dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
226 | ASSERT3U(ISP2(dn->dn_datablksz), ==, dn->dn_datablkshift != 0); | |
227 | ASSERT3U((dn->dn_nblkptr - 1) * sizeof (blkptr_t) + | |
228 | dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); | |
229 | for (i = 0; i < TXG_SIZE; i++) { | |
230 | ASSERT3U(dn->dn_next_nlevels[i], <=, dn->dn_nlevels); | |
231 | } | |
232 | } | |
233 | if (dn->dn_phys->dn_type != DMU_OT_NONE) | |
234 | ASSERT3U(dn->dn_phys->dn_nlevels, <=, dn->dn_nlevels); | |
9babb374 | 235 | ASSERT(DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_dbuf != NULL); |
34dc7c2f BB |
236 | if (dn->dn_dbuf != NULL) { |
237 | ASSERT3P(dn->dn_phys, ==, | |
238 | (dnode_phys_t *)dn->dn_dbuf->db.db_data + | |
239 | (dn->dn_object % (dn->dn_dbuf->db.db_size >> DNODE_SHIFT))); | |
240 | } | |
241 | if (drop_struct_lock) | |
242 | rw_exit(&dn->dn_struct_rwlock); | |
243 | } | |
244 | #endif | |
245 | ||
246 | void | |
247 | dnode_byteswap(dnode_phys_t *dnp) | |
248 | { | |
249 | uint64_t *buf64 = (void*)&dnp->dn_blkptr; | |
250 | int i; | |
251 | ||
252 | if (dnp->dn_type == DMU_OT_NONE) { | |
253 | bzero(dnp, sizeof (dnode_phys_t)); | |
254 | return; | |
255 | } | |
256 | ||
257 | dnp->dn_datablkszsec = BSWAP_16(dnp->dn_datablkszsec); | |
258 | dnp->dn_bonuslen = BSWAP_16(dnp->dn_bonuslen); | |
259 | dnp->dn_maxblkid = BSWAP_64(dnp->dn_maxblkid); | |
260 | dnp->dn_used = BSWAP_64(dnp->dn_used); | |
261 | ||
262 | /* | |
263 | * dn_nblkptr is only one byte, so it's OK to read it in either | |
264 | * byte order. We can't read dn_bouslen. | |
265 | */ | |
266 | ASSERT(dnp->dn_indblkshift <= SPA_MAXBLOCKSHIFT); | |
267 | ASSERT(dnp->dn_nblkptr <= DN_MAX_NBLKPTR); | |
268 | for (i = 0; i < dnp->dn_nblkptr * sizeof (blkptr_t)/8; i++) | |
269 | buf64[i] = BSWAP_64(buf64[i]); | |
270 | ||
271 | /* | |
272 | * OK to check dn_bonuslen for zero, because it won't matter if | |
273 | * we have the wrong byte order. This is necessary because the | |
274 | * dnode dnode is smaller than a regular dnode. | |
275 | */ | |
276 | if (dnp->dn_bonuslen != 0) { | |
277 | /* | |
278 | * Note that the bonus length calculated here may be | |
279 | * longer than the actual bonus buffer. This is because | |
280 | * we always put the bonus buffer after the last block | |
281 | * pointer (instead of packing it against the end of the | |
282 | * dnode buffer). | |
283 | */ | |
284 | int off = (dnp->dn_nblkptr-1) * sizeof (blkptr_t); | |
285 | size_t len = DN_MAX_BONUSLEN - off; | |
9ae529ec CS |
286 | dmu_object_byteswap_t byteswap; |
287 | ASSERT(DMU_OT_IS_VALID(dnp->dn_bonustype)); | |
288 | byteswap = DMU_OT_BYTESWAP(dnp->dn_bonustype); | |
289 | dmu_ot_byteswap[byteswap].ob_func(dnp->dn_bonus + off, len); | |
34dc7c2f | 290 | } |
428870ff BB |
291 | |
292 | /* Swap SPILL block if we have one */ | |
293 | if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) | |
294 | byteswap_uint64_array(&dnp->dn_spill, sizeof (blkptr_t)); | |
295 | ||
34dc7c2f BB |
296 | } |
297 | ||
298 | void | |
299 | dnode_buf_byteswap(void *vbuf, size_t size) | |
300 | { | |
301 | dnode_phys_t *buf = vbuf; | |
302 | int i; | |
303 | ||
304 | ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT)); | |
305 | ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0); | |
306 | ||
307 | size >>= DNODE_SHIFT; | |
308 | for (i = 0; i < size; i++) { | |
309 | dnode_byteswap(buf); | |
310 | buf++; | |
311 | } | |
312 | } | |
313 | ||
34dc7c2f BB |
314 | void |
315 | dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx) | |
316 | { | |
317 | ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); | |
318 | ||
319 | dnode_setdirty(dn, tx); | |
320 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
321 | ASSERT3U(newsize, <=, DN_MAX_BONUSLEN - | |
322 | (dn->dn_nblkptr-1) * sizeof (blkptr_t)); | |
323 | dn->dn_bonuslen = newsize; | |
324 | if (newsize == 0) | |
325 | dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = DN_ZERO_BONUSLEN; | |
326 | else | |
327 | dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; | |
328 | rw_exit(&dn->dn_struct_rwlock); | |
329 | } | |
330 | ||
428870ff BB |
331 | void |
332 | dnode_setbonus_type(dnode_t *dn, dmu_object_type_t newtype, dmu_tx_t *tx) | |
333 | { | |
334 | ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); | |
335 | dnode_setdirty(dn, tx); | |
336 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
337 | dn->dn_bonustype = newtype; | |
338 | dn->dn_next_bonustype[tx->tx_txg & TXG_MASK] = dn->dn_bonustype; | |
339 | rw_exit(&dn->dn_struct_rwlock); | |
340 | } | |
341 | ||
342 | void | |
343 | dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx) | |
344 | { | |
345 | ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); | |
346 | ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); | |
347 | dnode_setdirty(dn, tx); | |
348 | dn->dn_rm_spillblk[tx->tx_txg&TXG_MASK] = DN_KILL_SPILLBLK; | |
349 | dn->dn_have_spill = B_FALSE; | |
350 | } | |
351 | ||
34dc7c2f BB |
352 | static void |
353 | dnode_setdblksz(dnode_t *dn, int size) | |
354 | { | |
c99c9001 | 355 | ASSERT0(P2PHASE(size, SPA_MINBLOCKSIZE)); |
34dc7c2f BB |
356 | ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); |
357 | ASSERT3U(size, >=, SPA_MINBLOCKSIZE); | |
358 | ASSERT3U(size >> SPA_MINBLOCKSHIFT, <, | |
359 | 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8)); | |
360 | dn->dn_datablksz = size; | |
361 | dn->dn_datablkszsec = size >> SPA_MINBLOCKSHIFT; | |
9bd274dd | 362 | dn->dn_datablkshift = ISP2(size) ? highbit64(size - 1) : 0; |
34dc7c2f BB |
363 | } |
364 | ||
365 | static dnode_t * | |
428870ff | 366 | dnode_create(objset_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, |
572e2857 | 367 | uint64_t object, dnode_handle_t *dnh) |
34dc7c2f | 368 | { |
79c76d5b | 369 | dnode_t *dn = kmem_cache_alloc(dnode_cache, KM_SLEEP); |
34dc7c2f | 370 | |
572e2857 BB |
371 | ASSERT(!POINTER_IS_VALID(dn->dn_objset)); |
372 | dn->dn_moved = 0; | |
373 | ||
374 | /* | |
375 | * Defer setting dn_objset until the dnode is ready to be a candidate | |
376 | * for the dnode_move() callback. | |
377 | */ | |
34dc7c2f BB |
378 | dn->dn_object = object; |
379 | dn->dn_dbuf = db; | |
572e2857 | 380 | dn->dn_handle = dnh; |
34dc7c2f BB |
381 | dn->dn_phys = dnp; |
382 | ||
572e2857 | 383 | if (dnp->dn_datablkszsec) { |
34dc7c2f | 384 | dnode_setdblksz(dn, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); |
572e2857 BB |
385 | } else { |
386 | dn->dn_datablksz = 0; | |
387 | dn->dn_datablkszsec = 0; | |
388 | dn->dn_datablkshift = 0; | |
389 | } | |
34dc7c2f BB |
390 | dn->dn_indblkshift = dnp->dn_indblkshift; |
391 | dn->dn_nlevels = dnp->dn_nlevels; | |
392 | dn->dn_type = dnp->dn_type; | |
393 | dn->dn_nblkptr = dnp->dn_nblkptr; | |
394 | dn->dn_checksum = dnp->dn_checksum; | |
395 | dn->dn_compress = dnp->dn_compress; | |
396 | dn->dn_bonustype = dnp->dn_bonustype; | |
397 | dn->dn_bonuslen = dnp->dn_bonuslen; | |
398 | dn->dn_maxblkid = dnp->dn_maxblkid; | |
428870ff BB |
399 | dn->dn_have_spill = ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0); |
400 | dn->dn_id_flags = 0; | |
34dc7c2f BB |
401 | |
402 | dmu_zfetch_init(&dn->dn_zfetch, dn); | |
403 | ||
9ae529ec | 404 | ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type)); |
572e2857 | 405 | |
34dc7c2f BB |
406 | mutex_enter(&os->os_lock); |
407 | list_insert_head(&os->os_dnodes, dn); | |
572e2857 BB |
408 | membar_producer(); |
409 | /* | |
410 | * Everything else must be valid before assigning dn_objset makes the | |
411 | * dnode eligible for dnode_move(). | |
412 | */ | |
413 | dn->dn_objset = os; | |
34dc7c2f BB |
414 | mutex_exit(&os->os_lock); |
415 | ||
d164b209 | 416 | arc_space_consume(sizeof (dnode_t), ARC_SPACE_OTHER); |
34dc7c2f BB |
417 | return (dn); |
418 | } | |
419 | ||
572e2857 BB |
420 | /* |
421 | * Caller must be holding the dnode handle, which is released upon return. | |
422 | */ | |
34dc7c2f BB |
423 | static void |
424 | dnode_destroy(dnode_t *dn) | |
425 | { | |
428870ff | 426 | objset_t *os = dn->dn_objset; |
34dc7c2f | 427 | |
428870ff | 428 | ASSERT((dn->dn_id_flags & DN_ID_NEW_EXIST) == 0); |
34dc7c2f BB |
429 | |
430 | mutex_enter(&os->os_lock); | |
572e2857 | 431 | POINTER_INVALIDATE(&dn->dn_objset); |
34dc7c2f BB |
432 | list_remove(&os->os_dnodes, dn); |
433 | mutex_exit(&os->os_lock); | |
434 | ||
572e2857 BB |
435 | /* the dnode can no longer move, so we can release the handle */ |
436 | zrl_remove(&dn->dn_handle->dnh_zrlock); | |
437 | ||
438 | dn->dn_allocated_txg = 0; | |
439 | dn->dn_free_txg = 0; | |
440 | dn->dn_assigned_txg = 0; | |
441 | ||
442 | dn->dn_dirtyctx = 0; | |
443 | if (dn->dn_dirtyctx_firstset != NULL) { | |
34dc7c2f BB |
444 | kmem_free(dn->dn_dirtyctx_firstset, 1); |
445 | dn->dn_dirtyctx_firstset = NULL; | |
446 | } | |
572e2857 | 447 | if (dn->dn_bonus != NULL) { |
34dc7c2f BB |
448 | mutex_enter(&dn->dn_bonus->db_mtx); |
449 | dbuf_evict(dn->dn_bonus); | |
450 | dn->dn_bonus = NULL; | |
451 | } | |
572e2857 BB |
452 | dn->dn_zio = NULL; |
453 | ||
454 | dn->dn_have_spill = B_FALSE; | |
455 | dn->dn_oldused = 0; | |
456 | dn->dn_oldflags = 0; | |
457 | dn->dn_olduid = 0; | |
458 | dn->dn_oldgid = 0; | |
459 | dn->dn_newuid = 0; | |
460 | dn->dn_newgid = 0; | |
461 | dn->dn_id_flags = 0; | |
b663a23d | 462 | dn->dn_unlisted_l0_blkid = 0; |
572e2857 BB |
463 | |
464 | dmu_zfetch_rele(&dn->dn_zfetch); | |
34dc7c2f | 465 | kmem_cache_free(dnode_cache, dn); |
d164b209 | 466 | arc_space_return(sizeof (dnode_t), ARC_SPACE_OTHER); |
34dc7c2f BB |
467 | } |
468 | ||
469 | void | |
470 | dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, | |
471 | dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) | |
472 | { | |
473 | int i; | |
474 | ||
475 | if (blocksize == 0) | |
476 | blocksize = 1 << zfs_default_bs; | |
477 | else if (blocksize > SPA_MAXBLOCKSIZE) | |
478 | blocksize = SPA_MAXBLOCKSIZE; | |
479 | else | |
480 | blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); | |
481 | ||
482 | if (ibs == 0) | |
483 | ibs = zfs_default_ibs; | |
484 | ||
485 | ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT); | |
486 | ||
487 | dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn->dn_objset, | |
488 | dn->dn_object, tx->tx_txg, blocksize, ibs); | |
489 | ||
490 | ASSERT(dn->dn_type == DMU_OT_NONE); | |
491 | ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); | |
492 | ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); | |
493 | ASSERT(ot != DMU_OT_NONE); | |
9ae529ec | 494 | ASSERT(DMU_OT_IS_VALID(ot)); |
34dc7c2f | 495 | ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || |
428870ff | 496 | (bonustype == DMU_OT_SA && bonuslen == 0) || |
34dc7c2f | 497 | (bonustype != DMU_OT_NONE && bonuslen != 0)); |
9ae529ec | 498 | ASSERT(DMU_OT_IS_VALID(bonustype)); |
34dc7c2f BB |
499 | ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); |
500 | ASSERT(dn->dn_type == DMU_OT_NONE); | |
c99c9001 MS |
501 | ASSERT0(dn->dn_maxblkid); |
502 | ASSERT0(dn->dn_allocated_txg); | |
503 | ASSERT0(dn->dn_assigned_txg); | |
34dc7c2f BB |
504 | ASSERT(refcount_is_zero(&dn->dn_tx_holds)); |
505 | ASSERT3U(refcount_count(&dn->dn_holds), <=, 1); | |
506 | ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); | |
507 | ||
508 | for (i = 0; i < TXG_SIZE; i++) { | |
c99c9001 MS |
509 | ASSERT0(dn->dn_next_nblkptr[i]); |
510 | ASSERT0(dn->dn_next_nlevels[i]); | |
511 | ASSERT0(dn->dn_next_indblkshift[i]); | |
512 | ASSERT0(dn->dn_next_bonuslen[i]); | |
513 | ASSERT0(dn->dn_next_bonustype[i]); | |
514 | ASSERT0(dn->dn_rm_spillblk[i]); | |
515 | ASSERT0(dn->dn_next_blksz[i]); | |
34dc7c2f BB |
516 | ASSERT(!list_link_active(&dn->dn_dirty_link[i])); |
517 | ASSERT3P(list_head(&dn->dn_dirty_records[i]), ==, NULL); | |
9bd274dd | 518 | ASSERT3P(dn->dn_free_ranges[i], ==, NULL); |
34dc7c2f BB |
519 | } |
520 | ||
521 | dn->dn_type = ot; | |
522 | dnode_setdblksz(dn, blocksize); | |
523 | dn->dn_indblkshift = ibs; | |
524 | dn->dn_nlevels = 1; | |
428870ff BB |
525 | if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */ |
526 | dn->dn_nblkptr = 1; | |
527 | else | |
528 | dn->dn_nblkptr = 1 + | |
529 | ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); | |
34dc7c2f BB |
530 | dn->dn_bonustype = bonustype; |
531 | dn->dn_bonuslen = bonuslen; | |
532 | dn->dn_checksum = ZIO_CHECKSUM_INHERIT; | |
533 | dn->dn_compress = ZIO_COMPRESS_INHERIT; | |
534 | dn->dn_dirtyctx = 0; | |
535 | ||
536 | dn->dn_free_txg = 0; | |
537 | if (dn->dn_dirtyctx_firstset) { | |
538 | kmem_free(dn->dn_dirtyctx_firstset, 1); | |
539 | dn->dn_dirtyctx_firstset = NULL; | |
540 | } | |
541 | ||
542 | dn->dn_allocated_txg = tx->tx_txg; | |
428870ff | 543 | dn->dn_id_flags = 0; |
34dc7c2f BB |
544 | |
545 | dnode_setdirty(dn, tx); | |
546 | dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs; | |
547 | dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; | |
428870ff | 548 | dn->dn_next_bonustype[tx->tx_txg & TXG_MASK] = dn->dn_bonustype; |
34dc7c2f BB |
549 | dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz; |
550 | } | |
551 | ||
552 | void | |
553 | dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, | |
554 | dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) | |
555 | { | |
9babb374 | 556 | int nblkptr; |
34dc7c2f BB |
557 | |
558 | ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); | |
559 | ASSERT3U(blocksize, <=, SPA_MAXBLOCKSIZE); | |
c99c9001 | 560 | ASSERT0(blocksize % SPA_MINBLOCKSIZE); |
34dc7c2f BB |
561 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); |
562 | ASSERT(tx->tx_txg != 0); | |
563 | ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || | |
428870ff BB |
564 | (bonustype != DMU_OT_NONE && bonuslen != 0) || |
565 | (bonustype == DMU_OT_SA && bonuslen == 0)); | |
9ae529ec | 566 | ASSERT(DMU_OT_IS_VALID(bonustype)); |
34dc7c2f BB |
567 | ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); |
568 | ||
34dc7c2f BB |
569 | /* clean up any unreferenced dbufs */ |
570 | dnode_evict_dbufs(dn); | |
d164b209 | 571 | |
428870ff BB |
572 | dn->dn_id_flags = 0; |
573 | ||
34dc7c2f | 574 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); |
34dc7c2f | 575 | dnode_setdirty(dn, tx); |
9babb374 BB |
576 | if (dn->dn_datablksz != blocksize) { |
577 | /* change blocksize */ | |
578 | ASSERT(dn->dn_maxblkid == 0 && | |
579 | (BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || | |
580 | dnode_block_freed(dn, 0))); | |
581 | dnode_setdblksz(dn, blocksize); | |
582 | dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize; | |
583 | } | |
584 | if (dn->dn_bonuslen != bonuslen) | |
585 | dn->dn_next_bonuslen[tx->tx_txg&TXG_MASK] = bonuslen; | |
428870ff BB |
586 | |
587 | if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */ | |
588 | nblkptr = 1; | |
589 | else | |
590 | nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); | |
591 | if (dn->dn_bonustype != bonustype) | |
592 | dn->dn_next_bonustype[tx->tx_txg&TXG_MASK] = bonustype; | |
d164b209 BB |
593 | if (dn->dn_nblkptr != nblkptr) |
594 | dn->dn_next_nblkptr[tx->tx_txg&TXG_MASK] = nblkptr; | |
428870ff BB |
595 | if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { |
596 | dbuf_rm_spill(dn, tx); | |
597 | dnode_rm_spill(dn, tx); | |
598 | } | |
34dc7c2f | 599 | rw_exit(&dn->dn_struct_rwlock); |
34dc7c2f BB |
600 | |
601 | /* change type */ | |
602 | dn->dn_type = ot; | |
603 | ||
604 | /* change bonus size and type */ | |
605 | mutex_enter(&dn->dn_mtx); | |
34dc7c2f BB |
606 | dn->dn_bonustype = bonustype; |
607 | dn->dn_bonuslen = bonuslen; | |
d164b209 | 608 | dn->dn_nblkptr = nblkptr; |
34dc7c2f BB |
609 | dn->dn_checksum = ZIO_CHECKSUM_INHERIT; |
610 | dn->dn_compress = ZIO_COMPRESS_INHERIT; | |
611 | ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); | |
612 | ||
d164b209 BB |
613 | /* fix up the bonus db_size */ |
614 | if (dn->dn_bonus) { | |
34dc7c2f BB |
615 | dn->dn_bonus->db.db_size = |
616 | DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t); | |
617 | ASSERT(dn->dn_bonuslen <= dn->dn_bonus->db.db_size); | |
618 | } | |
619 | ||
620 | dn->dn_allocated_txg = tx->tx_txg; | |
621 | mutex_exit(&dn->dn_mtx); | |
622 | } | |
623 | ||
5ac1241a | 624 | #ifdef _KERNEL |
572e2857 BB |
625 | #ifdef DNODE_STATS |
626 | static struct { | |
627 | uint64_t dms_dnode_invalid; | |
628 | uint64_t dms_dnode_recheck1; | |
629 | uint64_t dms_dnode_recheck2; | |
630 | uint64_t dms_dnode_special; | |
631 | uint64_t dms_dnode_handle; | |
632 | uint64_t dms_dnode_rwlock; | |
633 | uint64_t dms_dnode_active; | |
634 | } dnode_move_stats; | |
635 | #endif /* DNODE_STATS */ | |
636 | ||
637 | static void | |
638 | dnode_move_impl(dnode_t *odn, dnode_t *ndn) | |
639 | { | |
640 | int i; | |
641 | ||
642 | ASSERT(!RW_LOCK_HELD(&odn->dn_struct_rwlock)); | |
643 | ASSERT(MUTEX_NOT_HELD(&odn->dn_mtx)); | |
644 | ASSERT(MUTEX_NOT_HELD(&odn->dn_dbufs_mtx)); | |
645 | ASSERT(!RW_LOCK_HELD(&odn->dn_zfetch.zf_rwlock)); | |
646 | ||
647 | /* Copy fields. */ | |
648 | ndn->dn_objset = odn->dn_objset; | |
649 | ndn->dn_object = odn->dn_object; | |
650 | ndn->dn_dbuf = odn->dn_dbuf; | |
651 | ndn->dn_handle = odn->dn_handle; | |
652 | ndn->dn_phys = odn->dn_phys; | |
653 | ndn->dn_type = odn->dn_type; | |
654 | ndn->dn_bonuslen = odn->dn_bonuslen; | |
655 | ndn->dn_bonustype = odn->dn_bonustype; | |
656 | ndn->dn_nblkptr = odn->dn_nblkptr; | |
657 | ndn->dn_checksum = odn->dn_checksum; | |
658 | ndn->dn_compress = odn->dn_compress; | |
659 | ndn->dn_nlevels = odn->dn_nlevels; | |
660 | ndn->dn_indblkshift = odn->dn_indblkshift; | |
661 | ndn->dn_datablkshift = odn->dn_datablkshift; | |
662 | ndn->dn_datablkszsec = odn->dn_datablkszsec; | |
663 | ndn->dn_datablksz = odn->dn_datablksz; | |
664 | ndn->dn_maxblkid = odn->dn_maxblkid; | |
665 | bcopy(&odn->dn_next_nblkptr[0], &ndn->dn_next_nblkptr[0], | |
666 | sizeof (odn->dn_next_nblkptr)); | |
667 | bcopy(&odn->dn_next_nlevels[0], &ndn->dn_next_nlevels[0], | |
668 | sizeof (odn->dn_next_nlevels)); | |
669 | bcopy(&odn->dn_next_indblkshift[0], &ndn->dn_next_indblkshift[0], | |
670 | sizeof (odn->dn_next_indblkshift)); | |
671 | bcopy(&odn->dn_next_bonustype[0], &ndn->dn_next_bonustype[0], | |
672 | sizeof (odn->dn_next_bonustype)); | |
673 | bcopy(&odn->dn_rm_spillblk[0], &ndn->dn_rm_spillblk[0], | |
674 | sizeof (odn->dn_rm_spillblk)); | |
675 | bcopy(&odn->dn_next_bonuslen[0], &ndn->dn_next_bonuslen[0], | |
676 | sizeof (odn->dn_next_bonuslen)); | |
677 | bcopy(&odn->dn_next_blksz[0], &ndn->dn_next_blksz[0], | |
678 | sizeof (odn->dn_next_blksz)); | |
679 | for (i = 0; i < TXG_SIZE; i++) { | |
680 | list_move_tail(&ndn->dn_dirty_records[i], | |
681 | &odn->dn_dirty_records[i]); | |
682 | } | |
9bd274dd MA |
683 | bcopy(&odn->dn_free_ranges[0], &ndn->dn_free_ranges[0], |
684 | sizeof (odn->dn_free_ranges)); | |
572e2857 BB |
685 | ndn->dn_allocated_txg = odn->dn_allocated_txg; |
686 | ndn->dn_free_txg = odn->dn_free_txg; | |
687 | ndn->dn_assigned_txg = odn->dn_assigned_txg; | |
688 | ndn->dn_dirtyctx = odn->dn_dirtyctx; | |
689 | ndn->dn_dirtyctx_firstset = odn->dn_dirtyctx_firstset; | |
690 | ASSERT(refcount_count(&odn->dn_tx_holds) == 0); | |
691 | refcount_transfer(&ndn->dn_holds, &odn->dn_holds); | |
692 | ASSERT(list_is_empty(&ndn->dn_dbufs)); | |
693 | list_move_tail(&ndn->dn_dbufs, &odn->dn_dbufs); | |
694 | ndn->dn_dbufs_count = odn->dn_dbufs_count; | |
b663a23d | 695 | ndn->dn_unlisted_l0_blkid = odn->dn_unlisted_l0_blkid; |
572e2857 BB |
696 | ndn->dn_bonus = odn->dn_bonus; |
697 | ndn->dn_have_spill = odn->dn_have_spill; | |
698 | ndn->dn_zio = odn->dn_zio; | |
699 | ndn->dn_oldused = odn->dn_oldused; | |
700 | ndn->dn_oldflags = odn->dn_oldflags; | |
701 | ndn->dn_olduid = odn->dn_olduid; | |
702 | ndn->dn_oldgid = odn->dn_oldgid; | |
703 | ndn->dn_newuid = odn->dn_newuid; | |
704 | ndn->dn_newgid = odn->dn_newgid; | |
705 | ndn->dn_id_flags = odn->dn_id_flags; | |
706 | dmu_zfetch_init(&ndn->dn_zfetch, NULL); | |
707 | list_move_tail(&ndn->dn_zfetch.zf_stream, &odn->dn_zfetch.zf_stream); | |
708 | ndn->dn_zfetch.zf_dnode = odn->dn_zfetch.zf_dnode; | |
709 | ndn->dn_zfetch.zf_stream_cnt = odn->dn_zfetch.zf_stream_cnt; | |
710 | ndn->dn_zfetch.zf_alloc_fail = odn->dn_zfetch.zf_alloc_fail; | |
711 | ||
712 | /* | |
713 | * Update back pointers. Updating the handle fixes the back pointer of | |
714 | * every descendant dbuf as well as the bonus dbuf. | |
715 | */ | |
716 | ASSERT(ndn->dn_handle->dnh_dnode == odn); | |
717 | ndn->dn_handle->dnh_dnode = ndn; | |
718 | if (ndn->dn_zfetch.zf_dnode == odn) { | |
719 | ndn->dn_zfetch.zf_dnode = ndn; | |
720 | } | |
721 | ||
722 | /* | |
723 | * Invalidate the original dnode by clearing all of its back pointers. | |
724 | */ | |
725 | odn->dn_dbuf = NULL; | |
726 | odn->dn_handle = NULL; | |
727 | list_create(&odn->dn_dbufs, sizeof (dmu_buf_impl_t), | |
728 | offsetof(dmu_buf_impl_t, db_link)); | |
729 | odn->dn_dbufs_count = 0; | |
b663a23d | 730 | odn->dn_unlisted_l0_blkid = 0; |
572e2857 BB |
731 | odn->dn_bonus = NULL; |
732 | odn->dn_zfetch.zf_dnode = NULL; | |
733 | ||
734 | /* | |
735 | * Set the low bit of the objset pointer to ensure that dnode_move() | |
736 | * recognizes the dnode as invalid in any subsequent callback. | |
737 | */ | |
738 | POINTER_INVALIDATE(&odn->dn_objset); | |
739 | ||
740 | /* | |
741 | * Satisfy the destructor. | |
742 | */ | |
743 | for (i = 0; i < TXG_SIZE; i++) { | |
744 | list_create(&odn->dn_dirty_records[i], | |
745 | sizeof (dbuf_dirty_record_t), | |
746 | offsetof(dbuf_dirty_record_t, dr_dirty_node)); | |
9bd274dd | 747 | odn->dn_free_ranges[i] = NULL; |
572e2857 BB |
748 | odn->dn_next_nlevels[i] = 0; |
749 | odn->dn_next_indblkshift[i] = 0; | |
750 | odn->dn_next_bonustype[i] = 0; | |
751 | odn->dn_rm_spillblk[i] = 0; | |
752 | odn->dn_next_bonuslen[i] = 0; | |
753 | odn->dn_next_blksz[i] = 0; | |
754 | } | |
755 | odn->dn_allocated_txg = 0; | |
756 | odn->dn_free_txg = 0; | |
757 | odn->dn_assigned_txg = 0; | |
758 | odn->dn_dirtyctx = 0; | |
759 | odn->dn_dirtyctx_firstset = NULL; | |
760 | odn->dn_have_spill = B_FALSE; | |
761 | odn->dn_zio = NULL; | |
762 | odn->dn_oldused = 0; | |
763 | odn->dn_oldflags = 0; | |
764 | odn->dn_olduid = 0; | |
765 | odn->dn_oldgid = 0; | |
766 | odn->dn_newuid = 0; | |
767 | odn->dn_newgid = 0; | |
768 | odn->dn_id_flags = 0; | |
769 | ||
770 | /* | |
771 | * Mark the dnode. | |
772 | */ | |
773 | ndn->dn_moved = 1; | |
774 | odn->dn_moved = (uint8_t)-1; | |
775 | } | |
776 | ||
572e2857 BB |
777 | /*ARGSUSED*/ |
778 | static kmem_cbrc_t | |
779 | dnode_move(void *buf, void *newbuf, size_t size, void *arg) | |
780 | { | |
781 | dnode_t *odn = buf, *ndn = newbuf; | |
782 | objset_t *os; | |
783 | int64_t refcount; | |
784 | uint32_t dbufs; | |
785 | ||
786 | /* | |
787 | * The dnode is on the objset's list of known dnodes if the objset | |
788 | * pointer is valid. We set the low bit of the objset pointer when | |
789 | * freeing the dnode to invalidate it, and the memory patterns written | |
790 | * by kmem (baddcafe and deadbeef) set at least one of the two low bits. | |
791 | * A newly created dnode sets the objset pointer last of all to indicate | |
792 | * that the dnode is known and in a valid state to be moved by this | |
793 | * function. | |
794 | */ | |
795 | os = odn->dn_objset; | |
796 | if (!POINTER_IS_VALID(os)) { | |
797 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_invalid); | |
798 | return (KMEM_CBRC_DONT_KNOW); | |
799 | } | |
800 | ||
801 | /* | |
802 | * Ensure that the objset does not go away during the move. | |
803 | */ | |
804 | rw_enter(&os_lock, RW_WRITER); | |
805 | if (os != odn->dn_objset) { | |
806 | rw_exit(&os_lock); | |
807 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_recheck1); | |
808 | return (KMEM_CBRC_DONT_KNOW); | |
809 | } | |
810 | ||
811 | /* | |
812 | * If the dnode is still valid, then so is the objset. We know that no | |
813 | * valid objset can be freed while we hold os_lock, so we can safely | |
814 | * ensure that the objset remains in use. | |
815 | */ | |
816 | mutex_enter(&os->os_lock); | |
817 | ||
818 | /* | |
819 | * Recheck the objset pointer in case the dnode was removed just before | |
820 | * acquiring the lock. | |
821 | */ | |
822 | if (os != odn->dn_objset) { | |
823 | mutex_exit(&os->os_lock); | |
824 | rw_exit(&os_lock); | |
825 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_recheck2); | |
826 | return (KMEM_CBRC_DONT_KNOW); | |
827 | } | |
828 | ||
829 | /* | |
830 | * At this point we know that as long as we hold os->os_lock, the dnode | |
831 | * cannot be freed and fields within the dnode can be safely accessed. | |
832 | * The objset listing this dnode cannot go away as long as this dnode is | |
833 | * on its list. | |
834 | */ | |
835 | rw_exit(&os_lock); | |
836 | if (DMU_OBJECT_IS_SPECIAL(odn->dn_object)) { | |
837 | mutex_exit(&os->os_lock); | |
838 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_special); | |
839 | return (KMEM_CBRC_NO); | |
840 | } | |
841 | ASSERT(odn->dn_dbuf != NULL); /* only "special" dnodes have no parent */ | |
842 | ||
843 | /* | |
844 | * Lock the dnode handle to prevent the dnode from obtaining any new | |
845 | * holds. This also prevents the descendant dbufs and the bonus dbuf | |
846 | * from accessing the dnode, so that we can discount their holds. The | |
847 | * handle is safe to access because we know that while the dnode cannot | |
848 | * go away, neither can its handle. Once we hold dnh_zrlock, we can | |
849 | * safely move any dnode referenced only by dbufs. | |
850 | */ | |
851 | if (!zrl_tryenter(&odn->dn_handle->dnh_zrlock)) { | |
852 | mutex_exit(&os->os_lock); | |
853 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_handle); | |
854 | return (KMEM_CBRC_LATER); | |
855 | } | |
856 | ||
857 | /* | |
858 | * Ensure a consistent view of the dnode's holds and the dnode's dbufs. | |
859 | * We need to guarantee that there is a hold for every dbuf in order to | |
860 | * determine whether the dnode is actively referenced. Falsely matching | |
861 | * a dbuf to an active hold would lead to an unsafe move. It's possible | |
862 | * that a thread already having an active dnode hold is about to add a | |
863 | * dbuf, and we can't compare hold and dbuf counts while the add is in | |
864 | * progress. | |
865 | */ | |
866 | if (!rw_tryenter(&odn->dn_struct_rwlock, RW_WRITER)) { | |
867 | zrl_exit(&odn->dn_handle->dnh_zrlock); | |
868 | mutex_exit(&os->os_lock); | |
869 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_rwlock); | |
870 | return (KMEM_CBRC_LATER); | |
871 | } | |
872 | ||
873 | /* | |
874 | * A dbuf may be removed (evicted) without an active dnode hold. In that | |
875 | * case, the dbuf count is decremented under the handle lock before the | |
876 | * dbuf's hold is released. This order ensures that if we count the hold | |
877 | * after the dbuf is removed but before its hold is released, we will | |
878 | * treat the unmatched hold as active and exit safely. If we count the | |
879 | * hold before the dbuf is removed, the hold is discounted, and the | |
880 | * removal is blocked until the move completes. | |
881 | */ | |
882 | refcount = refcount_count(&odn->dn_holds); | |
883 | ASSERT(refcount >= 0); | |
884 | dbufs = odn->dn_dbufs_count; | |
885 | ||
886 | /* We can't have more dbufs than dnode holds. */ | |
887 | ASSERT3U(dbufs, <=, refcount); | |
888 | DTRACE_PROBE3(dnode__move, dnode_t *, odn, int64_t, refcount, | |
889 | uint32_t, dbufs); | |
890 | ||
891 | if (refcount > dbufs) { | |
892 | rw_exit(&odn->dn_struct_rwlock); | |
893 | zrl_exit(&odn->dn_handle->dnh_zrlock); | |
894 | mutex_exit(&os->os_lock); | |
895 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_active); | |
896 | return (KMEM_CBRC_LATER); | |
897 | } | |
898 | ||
899 | rw_exit(&odn->dn_struct_rwlock); | |
900 | ||
901 | /* | |
902 | * At this point we know that anyone with a hold on the dnode is not | |
903 | * actively referencing it. The dnode is known and in a valid state to | |
904 | * move. We're holding the locks needed to execute the critical section. | |
905 | */ | |
906 | dnode_move_impl(odn, ndn); | |
907 | ||
908 | list_link_replace(&odn->dn_link, &ndn->dn_link); | |
909 | /* If the dnode was safe to move, the refcount cannot have changed. */ | |
910 | ASSERT(refcount == refcount_count(&ndn->dn_holds)); | |
911 | ASSERT(dbufs == ndn->dn_dbufs_count); | |
912 | zrl_exit(&ndn->dn_handle->dnh_zrlock); /* handle has moved */ | |
913 | mutex_exit(&os->os_lock); | |
914 | ||
915 | return (KMEM_CBRC_YES); | |
916 | } | |
917 | #endif /* _KERNEL */ | |
918 | ||
34dc7c2f | 919 | void |
572e2857 | 920 | dnode_special_close(dnode_handle_t *dnh) |
34dc7c2f | 921 | { |
572e2857 BB |
922 | dnode_t *dn = dnh->dnh_dnode; |
923 | ||
34dc7c2f BB |
924 | /* |
925 | * Wait for final references to the dnode to clear. This can | |
926 | * only happen if the arc is asyncronously evicting state that | |
927 | * has a hold on this dnode while we are trying to evict this | |
928 | * dnode. | |
929 | */ | |
930 | while (refcount_count(&dn->dn_holds) > 0) | |
931 | delay(1); | |
572e2857 BB |
932 | zrl_add(&dnh->dnh_zrlock); |
933 | dnode_destroy(dn); /* implicit zrl_remove() */ | |
934 | zrl_destroy(&dnh->dnh_zrlock); | |
935 | dnh->dnh_dnode = NULL; | |
34dc7c2f BB |
936 | } |
937 | ||
938 | dnode_t * | |
572e2857 BB |
939 | dnode_special_open(objset_t *os, dnode_phys_t *dnp, uint64_t object, |
940 | dnode_handle_t *dnh) | |
34dc7c2f | 941 | { |
572e2857 BB |
942 | dnode_t *dn = dnode_create(os, dnp, NULL, object, dnh); |
943 | dnh->dnh_dnode = dn; | |
944 | zrl_init(&dnh->dnh_zrlock); | |
34dc7c2f BB |
945 | DNODE_VERIFY(dn); |
946 | return (dn); | |
947 | } | |
948 | ||
949 | static void | |
950 | dnode_buf_pageout(dmu_buf_t *db, void *arg) | |
951 | { | |
572e2857 | 952 | dnode_children_t *children_dnodes = arg; |
34dc7c2f BB |
953 | int i; |
954 | int epb = db->db_size >> DNODE_SHIFT; | |
955 | ||
572e2857 BB |
956 | ASSERT(epb == children_dnodes->dnc_count); |
957 | ||
34dc7c2f | 958 | for (i = 0; i < epb; i++) { |
572e2857 BB |
959 | dnode_handle_t *dnh = &children_dnodes->dnc_children[i]; |
960 | dnode_t *dn; | |
34dc7c2f | 961 | |
572e2857 BB |
962 | /* |
963 | * The dnode handle lock guards against the dnode moving to | |
964 | * another valid address, so there is no need here to guard | |
965 | * against changes to or from NULL. | |
966 | */ | |
967 | if (dnh->dnh_dnode == NULL) { | |
968 | zrl_destroy(&dnh->dnh_zrlock); | |
34dc7c2f | 969 | continue; |
572e2857 BB |
970 | } |
971 | ||
972 | zrl_add(&dnh->dnh_zrlock); | |
973 | dn = dnh->dnh_dnode; | |
34dc7c2f BB |
974 | /* |
975 | * If there are holds on this dnode, then there should | |
976 | * be holds on the dnode's containing dbuf as well; thus | |
572e2857 | 977 | * it wouldn't be eligible for eviction and this function |
34dc7c2f BB |
978 | * would not have been called. |
979 | */ | |
980 | ASSERT(refcount_is_zero(&dn->dn_holds)); | |
34dc7c2f BB |
981 | ASSERT(refcount_is_zero(&dn->dn_tx_holds)); |
982 | ||
572e2857 BB |
983 | dnode_destroy(dn); /* implicit zrl_remove() */ |
984 | zrl_destroy(&dnh->dnh_zrlock); | |
985 | dnh->dnh_dnode = NULL; | |
34dc7c2f | 986 | } |
572e2857 BB |
987 | kmem_free(children_dnodes, sizeof (dnode_children_t) + |
988 | (epb - 1) * sizeof (dnode_handle_t)); | |
34dc7c2f BB |
989 | } |
990 | ||
991 | /* | |
992 | * errors: | |
993 | * EINVAL - invalid object number. | |
994 | * EIO - i/o error. | |
995 | * succeeds even for free dnodes. | |
996 | */ | |
997 | int | |
428870ff | 998 | dnode_hold_impl(objset_t *os, uint64_t object, int flag, |
34dc7c2f BB |
999 | void *tag, dnode_t **dnp) |
1000 | { | |
1001 | int epb, idx, err; | |
1002 | int drop_struct_lock = FALSE; | |
1003 | int type; | |
1004 | uint64_t blk; | |
1005 | dnode_t *mdn, *dn; | |
1006 | dmu_buf_impl_t *db; | |
572e2857 BB |
1007 | dnode_children_t *children_dnodes; |
1008 | dnode_handle_t *dnh; | |
34dc7c2f | 1009 | |
b128c09f BB |
1010 | /* |
1011 | * If you are holding the spa config lock as writer, you shouldn't | |
428870ff BB |
1012 | * be asking the DMU to do *anything* unless it's the root pool |
1013 | * which may require us to read from the root filesystem while | |
1014 | * holding some (not all) of the locks as writer. | |
b128c09f | 1015 | */ |
428870ff BB |
1016 | ASSERT(spa_config_held(os->os_spa, SCL_ALL, RW_WRITER) == 0 || |
1017 | (spa_is_root(os->os_spa) && | |
572e2857 | 1018 | spa_config_held(os->os_spa, SCL_STATE, RW_WRITER))); |
b128c09f | 1019 | |
9babb374 BB |
1020 | if (object == DMU_USERUSED_OBJECT || object == DMU_GROUPUSED_OBJECT) { |
1021 | dn = (object == DMU_USERUSED_OBJECT) ? | |
572e2857 | 1022 | DMU_USERUSED_DNODE(os) : DMU_GROUPUSED_DNODE(os); |
9babb374 | 1023 | if (dn == NULL) |
2e528b49 | 1024 | return (SET_ERROR(ENOENT)); |
9babb374 BB |
1025 | type = dn->dn_type; |
1026 | if ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) | |
2e528b49 | 1027 | return (SET_ERROR(ENOENT)); |
9babb374 | 1028 | if ((flag & DNODE_MUST_BE_FREE) && type != DMU_OT_NONE) |
2e528b49 | 1029 | return (SET_ERROR(EEXIST)); |
9babb374 BB |
1030 | DNODE_VERIFY(dn); |
1031 | (void) refcount_add(&dn->dn_holds, tag); | |
1032 | *dnp = dn; | |
1033 | return (0); | |
1034 | } | |
1035 | ||
34dc7c2f | 1036 | if (object == 0 || object >= DN_MAX_OBJECT) |
2e528b49 | 1037 | return (SET_ERROR(EINVAL)); |
34dc7c2f | 1038 | |
572e2857 BB |
1039 | mdn = DMU_META_DNODE(os); |
1040 | ASSERT(mdn->dn_object == DMU_META_DNODE_OBJECT); | |
34dc7c2f BB |
1041 | |
1042 | DNODE_VERIFY(mdn); | |
1043 | ||
1044 | if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) { | |
1045 | rw_enter(&mdn->dn_struct_rwlock, RW_READER); | |
1046 | drop_struct_lock = TRUE; | |
1047 | } | |
1048 | ||
1049 | blk = dbuf_whichblock(mdn, object * sizeof (dnode_phys_t)); | |
1050 | ||
1051 | db = dbuf_hold(mdn, blk, FTAG); | |
1052 | if (drop_struct_lock) | |
1053 | rw_exit(&mdn->dn_struct_rwlock); | |
1054 | if (db == NULL) | |
2e528b49 | 1055 | return (SET_ERROR(EIO)); |
34dc7c2f BB |
1056 | err = dbuf_read(db, NULL, DB_RF_CANFAIL); |
1057 | if (err) { | |
1058 | dbuf_rele(db, FTAG); | |
1059 | return (err); | |
1060 | } | |
1061 | ||
1062 | ASSERT3U(db->db.db_size, >=, 1<<DNODE_SHIFT); | |
1063 | epb = db->db.db_size >> DNODE_SHIFT; | |
1064 | ||
1065 | idx = object & (epb-1); | |
1066 | ||
572e2857 | 1067 | ASSERT(DB_DNODE(db)->dn_type == DMU_OT_DNODE); |
34dc7c2f BB |
1068 | children_dnodes = dmu_buf_get_user(&db->db); |
1069 | if (children_dnodes == NULL) { | |
572e2857 BB |
1070 | int i; |
1071 | dnode_children_t *winner; | |
1072 | children_dnodes = kmem_alloc(sizeof (dnode_children_t) + | |
79c76d5b | 1073 | (epb - 1) * sizeof (dnode_handle_t), KM_SLEEP); |
572e2857 BB |
1074 | children_dnodes->dnc_count = epb; |
1075 | dnh = &children_dnodes->dnc_children[0]; | |
1076 | for (i = 0; i < epb; i++) { | |
1077 | zrl_init(&dnh[i].dnh_zrlock); | |
1078 | dnh[i].dnh_dnode = NULL; | |
1079 | } | |
c65aa5b2 BB |
1080 | if ((winner = dmu_buf_set_user(&db->db, children_dnodes, NULL, |
1081 | dnode_buf_pageout))) { | |
572e2857 BB |
1082 | kmem_free(children_dnodes, sizeof (dnode_children_t) + |
1083 | (epb - 1) * sizeof (dnode_handle_t)); | |
34dc7c2f BB |
1084 | children_dnodes = winner; |
1085 | } | |
1086 | } | |
572e2857 | 1087 | ASSERT(children_dnodes->dnc_count == epb); |
34dc7c2f | 1088 | |
572e2857 BB |
1089 | dnh = &children_dnodes->dnc_children[idx]; |
1090 | zrl_add(&dnh->dnh_zrlock); | |
1091 | if ((dn = dnh->dnh_dnode) == NULL) { | |
1092 | dnode_phys_t *phys = (dnode_phys_t *)db->db.db_data+idx; | |
34dc7c2f BB |
1093 | dnode_t *winner; |
1094 | ||
572e2857 BB |
1095 | dn = dnode_create(os, phys, db, object, dnh); |
1096 | winner = atomic_cas_ptr(&dnh->dnh_dnode, NULL, dn); | |
34dc7c2f | 1097 | if (winner != NULL) { |
572e2857 BB |
1098 | zrl_add(&dnh->dnh_zrlock); |
1099 | dnode_destroy(dn); /* implicit zrl_remove() */ | |
34dc7c2f BB |
1100 | dn = winner; |
1101 | } | |
1102 | } | |
1103 | ||
1104 | mutex_enter(&dn->dn_mtx); | |
1105 | type = dn->dn_type; | |
1106 | if (dn->dn_free_txg || | |
1107 | ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) || | |
9babb374 | 1108 | ((flag & DNODE_MUST_BE_FREE) && |
428870ff | 1109 | (type != DMU_OT_NONE || !refcount_is_zero(&dn->dn_holds)))) { |
34dc7c2f | 1110 | mutex_exit(&dn->dn_mtx); |
572e2857 | 1111 | zrl_remove(&dnh->dnh_zrlock); |
34dc7c2f BB |
1112 | dbuf_rele(db, FTAG); |
1113 | return (type == DMU_OT_NONE ? ENOENT : EEXIST); | |
1114 | } | |
1115 | mutex_exit(&dn->dn_mtx); | |
1116 | ||
1117 | if (refcount_add(&dn->dn_holds, tag) == 1) | |
572e2857 BB |
1118 | dbuf_add_ref(db, dnh); |
1119 | /* Now we can rely on the hold to prevent the dnode from moving. */ | |
1120 | zrl_remove(&dnh->dnh_zrlock); | |
34dc7c2f BB |
1121 | |
1122 | DNODE_VERIFY(dn); | |
1123 | ASSERT3P(dn->dn_dbuf, ==, db); | |
1124 | ASSERT3U(dn->dn_object, ==, object); | |
1125 | dbuf_rele(db, FTAG); | |
1126 | ||
1127 | *dnp = dn; | |
1128 | return (0); | |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * Return held dnode if the object is allocated, NULL if not. | |
1133 | */ | |
1134 | int | |
428870ff | 1135 | dnode_hold(objset_t *os, uint64_t object, void *tag, dnode_t **dnp) |
34dc7c2f BB |
1136 | { |
1137 | return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp)); | |
1138 | } | |
1139 | ||
1140 | /* | |
1141 | * Can only add a reference if there is already at least one | |
1142 | * reference on the dnode. Returns FALSE if unable to add a | |
1143 | * new reference. | |
1144 | */ | |
1145 | boolean_t | |
1146 | dnode_add_ref(dnode_t *dn, void *tag) | |
1147 | { | |
1148 | mutex_enter(&dn->dn_mtx); | |
1149 | if (refcount_is_zero(&dn->dn_holds)) { | |
1150 | mutex_exit(&dn->dn_mtx); | |
1151 | return (FALSE); | |
1152 | } | |
1153 | VERIFY(1 < refcount_add(&dn->dn_holds, tag)); | |
1154 | mutex_exit(&dn->dn_mtx); | |
1155 | return (TRUE); | |
1156 | } | |
1157 | ||
1158 | void | |
1159 | dnode_rele(dnode_t *dn, void *tag) | |
4c7b7eed JG |
1160 | { |
1161 | mutex_enter(&dn->dn_mtx); | |
1162 | dnode_rele_and_unlock(dn, tag); | |
1163 | } | |
1164 | ||
1165 | void | |
1166 | dnode_rele_and_unlock(dnode_t *dn, void *tag) | |
34dc7c2f BB |
1167 | { |
1168 | uint64_t refs; | |
572e2857 BB |
1169 | /* Get while the hold prevents the dnode from moving. */ |
1170 | dmu_buf_impl_t *db = dn->dn_dbuf; | |
1171 | dnode_handle_t *dnh = dn->dn_handle; | |
34dc7c2f | 1172 | |
34dc7c2f BB |
1173 | refs = refcount_remove(&dn->dn_holds, tag); |
1174 | mutex_exit(&dn->dn_mtx); | |
572e2857 BB |
1175 | |
1176 | /* | |
1177 | * It's unsafe to release the last hold on a dnode by dnode_rele() or | |
1178 | * indirectly by dbuf_rele() while relying on the dnode handle to | |
1179 | * prevent the dnode from moving, since releasing the last hold could | |
1180 | * result in the dnode's parent dbuf evicting its dnode handles. For | |
1181 | * that reason anyone calling dnode_rele() or dbuf_rele() without some | |
1182 | * other direct or indirect hold on the dnode must first drop the dnode | |
1183 | * handle. | |
1184 | */ | |
1185 | ASSERT(refs > 0 || dnh->dnh_zrlock.zr_owner != curthread); | |
1186 | ||
34dc7c2f | 1187 | /* NOTE: the DNODE_DNODE does not have a dn_dbuf */ |
572e2857 BB |
1188 | if (refs == 0 && db != NULL) { |
1189 | /* | |
1190 | * Another thread could add a hold to the dnode handle in | |
1191 | * dnode_hold_impl() while holding the parent dbuf. Since the | |
1192 | * hold on the parent dbuf prevents the handle from being | |
1193 | * destroyed, the hold on the handle is OK. We can't yet assert | |
1194 | * that the handle has zero references, but that will be | |
1195 | * asserted anyway when the handle gets destroyed. | |
1196 | */ | |
1197 | dbuf_rele(db, dnh); | |
1198 | } | |
34dc7c2f BB |
1199 | } |
1200 | ||
1201 | void | |
1202 | dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) | |
1203 | { | |
428870ff | 1204 | objset_t *os = dn->dn_objset; |
34dc7c2f BB |
1205 | uint64_t txg = tx->tx_txg; |
1206 | ||
9babb374 BB |
1207 | if (DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { |
1208 | dsl_dataset_dirty(os->os_dsl_dataset, tx); | |
34dc7c2f | 1209 | return; |
9babb374 | 1210 | } |
34dc7c2f BB |
1211 | |
1212 | DNODE_VERIFY(dn); | |
1213 | ||
1214 | #ifdef ZFS_DEBUG | |
1215 | mutex_enter(&dn->dn_mtx); | |
1216 | ASSERT(dn->dn_phys->dn_type || dn->dn_allocated_txg); | |
572e2857 | 1217 | ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); |
34dc7c2f BB |
1218 | mutex_exit(&dn->dn_mtx); |
1219 | #endif | |
1220 | ||
428870ff BB |
1221 | /* |
1222 | * Determine old uid/gid when necessary | |
1223 | */ | |
1224 | dmu_objset_userquota_get_ids(dn, B_TRUE, tx); | |
1225 | ||
34dc7c2f BB |
1226 | mutex_enter(&os->os_lock); |
1227 | ||
1228 | /* | |
1229 | * If we are already marked dirty, we're done. | |
1230 | */ | |
1231 | if (list_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) { | |
1232 | mutex_exit(&os->os_lock); | |
1233 | return; | |
1234 | } | |
1235 | ||
1236 | ASSERT(!refcount_is_zero(&dn->dn_holds) || list_head(&dn->dn_dbufs)); | |
1237 | ASSERT(dn->dn_datablksz != 0); | |
c99c9001 MS |
1238 | ASSERT0(dn->dn_next_bonuslen[txg&TXG_MASK]); |
1239 | ASSERT0(dn->dn_next_blksz[txg&TXG_MASK]); | |
1240 | ASSERT0(dn->dn_next_bonustype[txg&TXG_MASK]); | |
34dc7c2f BB |
1241 | |
1242 | dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n", | |
1243 | dn->dn_object, txg); | |
1244 | ||
1245 | if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) { | |
1246 | list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn); | |
1247 | } else { | |
1248 | list_insert_tail(&os->os_dirty_dnodes[txg&TXG_MASK], dn); | |
1249 | } | |
1250 | ||
1251 | mutex_exit(&os->os_lock); | |
1252 | ||
1253 | /* | |
1254 | * The dnode maintains a hold on its containing dbuf as | |
1255 | * long as there are holds on it. Each instantiated child | |
572e2857 | 1256 | * dbuf maintains a hold on the dnode. When the last child |
34dc7c2f BB |
1257 | * drops its hold, the dnode will drop its hold on the |
1258 | * containing dbuf. We add a "dirty hold" here so that the | |
1259 | * dnode will hang around after we finish processing its | |
1260 | * children. | |
1261 | */ | |
1262 | VERIFY(dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg)); | |
1263 | ||
1264 | (void) dbuf_dirty(dn->dn_dbuf, tx); | |
1265 | ||
1266 | dsl_dataset_dirty(os->os_dsl_dataset, tx); | |
1267 | } | |
1268 | ||
1269 | void | |
1270 | dnode_free(dnode_t *dn, dmu_tx_t *tx) | |
1271 | { | |
1272 | int txgoff = tx->tx_txg & TXG_MASK; | |
1273 | ||
1274 | dprintf("dn=%p txg=%llu\n", dn, tx->tx_txg); | |
1275 | ||
1276 | /* we should be the only holder... hopefully */ | |
1277 | /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */ | |
1278 | ||
1279 | mutex_enter(&dn->dn_mtx); | |
1280 | if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg) { | |
1281 | mutex_exit(&dn->dn_mtx); | |
1282 | return; | |
1283 | } | |
1284 | dn->dn_free_txg = tx->tx_txg; | |
1285 | mutex_exit(&dn->dn_mtx); | |
1286 | ||
1287 | /* | |
1288 | * If the dnode is already dirty, it needs to be moved from | |
1289 | * the dirty list to the free list. | |
1290 | */ | |
1291 | mutex_enter(&dn->dn_objset->os_lock); | |
1292 | if (list_link_active(&dn->dn_dirty_link[txgoff])) { | |
1293 | list_remove(&dn->dn_objset->os_dirty_dnodes[txgoff], dn); | |
1294 | list_insert_tail(&dn->dn_objset->os_free_dnodes[txgoff], dn); | |
1295 | mutex_exit(&dn->dn_objset->os_lock); | |
1296 | } else { | |
1297 | mutex_exit(&dn->dn_objset->os_lock); | |
1298 | dnode_setdirty(dn, tx); | |
1299 | } | |
1300 | } | |
1301 | ||
1302 | /* | |
1303 | * Try to change the block size for the indicated dnode. This can only | |
1304 | * succeed if there are no blocks allocated or dirty beyond first block | |
1305 | */ | |
1306 | int | |
1307 | dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) | |
1308 | { | |
1309 | dmu_buf_impl_t *db, *db_next; | |
b128c09f | 1310 | int err; |
34dc7c2f BB |
1311 | |
1312 | if (size == 0) | |
1313 | size = SPA_MINBLOCKSIZE; | |
1314 | if (size > SPA_MAXBLOCKSIZE) | |
1315 | size = SPA_MAXBLOCKSIZE; | |
1316 | else | |
1317 | size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); | |
1318 | ||
1319 | if (ibs == dn->dn_indblkshift) | |
1320 | ibs = 0; | |
1321 | ||
1322 | if (size >> SPA_MINBLOCKSHIFT == dn->dn_datablkszsec && ibs == 0) | |
1323 | return (0); | |
1324 | ||
1325 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1326 | ||
1327 | /* Check for any allocated blocks beyond the first */ | |
93cf2076 | 1328 | if (dn->dn_maxblkid != 0) |
34dc7c2f BB |
1329 | goto fail; |
1330 | ||
1331 | mutex_enter(&dn->dn_dbufs_mtx); | |
1332 | for (db = list_head(&dn->dn_dbufs); db; db = db_next) { | |
1333 | db_next = list_next(&dn->dn_dbufs, db); | |
1334 | ||
428870ff BB |
1335 | if (db->db_blkid != 0 && db->db_blkid != DMU_BONUS_BLKID && |
1336 | db->db_blkid != DMU_SPILL_BLKID) { | |
34dc7c2f BB |
1337 | mutex_exit(&dn->dn_dbufs_mtx); |
1338 | goto fail; | |
1339 | } | |
1340 | } | |
1341 | mutex_exit(&dn->dn_dbufs_mtx); | |
1342 | ||
1343 | if (ibs && dn->dn_nlevels != 1) | |
1344 | goto fail; | |
1345 | ||
b128c09f BB |
1346 | /* resize the old block */ |
1347 | err = dbuf_hold_impl(dn, 0, 0, TRUE, FTAG, &db); | |
1348 | if (err == 0) | |
34dc7c2f | 1349 | dbuf_new_size(db, size, tx); |
b128c09f BB |
1350 | else if (err != ENOENT) |
1351 | goto fail; | |
34dc7c2f BB |
1352 | |
1353 | dnode_setdblksz(dn, size); | |
1354 | dnode_setdirty(dn, tx); | |
1355 | dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size; | |
1356 | if (ibs) { | |
1357 | dn->dn_indblkshift = ibs; | |
1358 | dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; | |
1359 | } | |
b128c09f | 1360 | /* rele after we have fixed the blocksize in the dnode */ |
34dc7c2f BB |
1361 | if (db) |
1362 | dbuf_rele(db, FTAG); | |
1363 | ||
1364 | rw_exit(&dn->dn_struct_rwlock); | |
1365 | return (0); | |
1366 | ||
1367 | fail: | |
1368 | rw_exit(&dn->dn_struct_rwlock); | |
2e528b49 | 1369 | return (SET_ERROR(ENOTSUP)); |
34dc7c2f BB |
1370 | } |
1371 | ||
b128c09f | 1372 | /* read-holding callers must not rely on the lock being continuously held */ |
34dc7c2f | 1373 | void |
b128c09f | 1374 | dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t have_read) |
34dc7c2f BB |
1375 | { |
1376 | uint64_t txgoff = tx->tx_txg & TXG_MASK; | |
34dc7c2f BB |
1377 | int epbs, new_nlevels; |
1378 | uint64_t sz; | |
1379 | ||
428870ff | 1380 | ASSERT(blkid != DMU_BONUS_BLKID); |
34dc7c2f | 1381 | |
b128c09f BB |
1382 | ASSERT(have_read ? |
1383 | RW_READ_HELD(&dn->dn_struct_rwlock) : | |
1384 | RW_WRITE_HELD(&dn->dn_struct_rwlock)); | |
1385 | ||
1386 | /* | |
1387 | * if we have a read-lock, check to see if we need to do any work | |
1388 | * before upgrading to a write-lock. | |
1389 | */ | |
1390 | if (have_read) { | |
1391 | if (blkid <= dn->dn_maxblkid) | |
1392 | return; | |
1393 | ||
1394 | if (!rw_tryupgrade(&dn->dn_struct_rwlock)) { | |
1395 | rw_exit(&dn->dn_struct_rwlock); | |
1396 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1397 | } | |
34dc7c2f BB |
1398 | } |
1399 | ||
1400 | if (blkid <= dn->dn_maxblkid) | |
1401 | goto out; | |
1402 | ||
1403 | dn->dn_maxblkid = blkid; | |
1404 | ||
1405 | /* | |
1406 | * Compute the number of levels necessary to support the new maxblkid. | |
1407 | */ | |
1408 | new_nlevels = 1; | |
1409 | epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
1410 | for (sz = dn->dn_nblkptr; | |
1411 | sz <= blkid && sz >= dn->dn_nblkptr; sz <<= epbs) | |
1412 | new_nlevels++; | |
1413 | ||
1414 | if (new_nlevels > dn->dn_nlevels) { | |
1415 | int old_nlevels = dn->dn_nlevels; | |
1416 | dmu_buf_impl_t *db; | |
1417 | list_t *list; | |
1418 | dbuf_dirty_record_t *new, *dr, *dr_next; | |
1419 | ||
1420 | dn->dn_nlevels = new_nlevels; | |
1421 | ||
1422 | ASSERT3U(new_nlevels, >, dn->dn_next_nlevels[txgoff]); | |
1423 | dn->dn_next_nlevels[txgoff] = new_nlevels; | |
1424 | ||
1425 | /* dirty the left indirects */ | |
1426 | db = dbuf_hold_level(dn, old_nlevels, 0, FTAG); | |
428870ff | 1427 | ASSERT(db != NULL); |
34dc7c2f BB |
1428 | new = dbuf_dirty(db, tx); |
1429 | dbuf_rele(db, FTAG); | |
1430 | ||
1431 | /* transfer the dirty records to the new indirect */ | |
1432 | mutex_enter(&dn->dn_mtx); | |
1433 | mutex_enter(&new->dt.di.dr_mtx); | |
1434 | list = &dn->dn_dirty_records[txgoff]; | |
1435 | for (dr = list_head(list); dr; dr = dr_next) { | |
1436 | dr_next = list_next(&dn->dn_dirty_records[txgoff], dr); | |
1437 | if (dr->dr_dbuf->db_level != new_nlevels-1 && | |
428870ff BB |
1438 | dr->dr_dbuf->db_blkid != DMU_BONUS_BLKID && |
1439 | dr->dr_dbuf->db_blkid != DMU_SPILL_BLKID) { | |
34dc7c2f BB |
1440 | ASSERT(dr->dr_dbuf->db_level == old_nlevels-1); |
1441 | list_remove(&dn->dn_dirty_records[txgoff], dr); | |
1442 | list_insert_tail(&new->dt.di.dr_children, dr); | |
1443 | dr->dr_parent = new; | |
1444 | } | |
1445 | } | |
1446 | mutex_exit(&new->dt.di.dr_mtx); | |
1447 | mutex_exit(&dn->dn_mtx); | |
1448 | } | |
1449 | ||
1450 | out: | |
b128c09f BB |
1451 | if (have_read) |
1452 | rw_downgrade(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
1453 | } |
1454 | ||
34dc7c2f BB |
1455 | void |
1456 | dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx) | |
1457 | { | |
1458 | dmu_buf_impl_t *db; | |
1459 | uint64_t blkoff, blkid, nblks; | |
b128c09f | 1460 | int blksz, blkshift, head, tail; |
34dc7c2f | 1461 | int trunc = FALSE; |
b128c09f | 1462 | int epbs; |
34dc7c2f BB |
1463 | |
1464 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1465 | blksz = dn->dn_datablksz; | |
b128c09f BB |
1466 | blkshift = dn->dn_datablkshift; |
1467 | epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
34dc7c2f | 1468 | |
b663a23d | 1469 | if (len == DMU_OBJECT_END) { |
34dc7c2f BB |
1470 | len = UINT64_MAX - off; |
1471 | trunc = TRUE; | |
1472 | } | |
1473 | ||
1474 | /* | |
1475 | * First, block align the region to free: | |
1476 | */ | |
1477 | if (ISP2(blksz)) { | |
1478 | head = P2NPHASE(off, blksz); | |
1479 | blkoff = P2PHASE(off, blksz); | |
b128c09f BB |
1480 | if ((off >> blkshift) > dn->dn_maxblkid) |
1481 | goto out; | |
34dc7c2f BB |
1482 | } else { |
1483 | ASSERT(dn->dn_maxblkid == 0); | |
1484 | if (off == 0 && len >= blksz) { | |
b0bc7a84 MG |
1485 | /* |
1486 | * Freeing the whole block; fast-track this request. | |
1487 | * Note that we won't dirty any indirect blocks, | |
1488 | * which is fine because we will be freeing the entire | |
1489 | * file and thus all indirect blocks will be freed | |
1490 | * by free_children(). | |
1491 | */ | |
b128c09f BB |
1492 | blkid = 0; |
1493 | nblks = 1; | |
1494 | goto done; | |
1495 | } else if (off >= blksz) { | |
1496 | /* Freeing past end-of-data */ | |
1497 | goto out; | |
34dc7c2f BB |
1498 | } else { |
1499 | /* Freeing part of the block. */ | |
1500 | head = blksz - off; | |
1501 | ASSERT3U(head, >, 0); | |
1502 | } | |
1503 | blkoff = off; | |
1504 | } | |
1505 | /* zero out any partial block data at the start of the range */ | |
1506 | if (head) { | |
1507 | ASSERT3U(blkoff + head, ==, blksz); | |
1508 | if (len < head) | |
1509 | head = len; | |
1510 | if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off), TRUE, | |
1511 | FTAG, &db) == 0) { | |
1512 | caddr_t data; | |
1513 | ||
1514 | /* don't dirty if it isn't on disk and isn't dirty */ | |
1515 | if (db->db_last_dirty || | |
1516 | (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { | |
1517 | rw_exit(&dn->dn_struct_rwlock); | |
b0bc7a84 | 1518 | dmu_buf_will_dirty(&db->db, tx); |
34dc7c2f BB |
1519 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); |
1520 | data = db->db.db_data; | |
1521 | bzero(data + blkoff, head); | |
1522 | } | |
1523 | dbuf_rele(db, FTAG); | |
1524 | } | |
1525 | off += head; | |
1526 | len -= head; | |
1527 | } | |
1528 | ||
1529 | /* If the range was less than one block, we're done */ | |
b128c09f | 1530 | if (len == 0) |
34dc7c2f BB |
1531 | goto out; |
1532 | ||
b128c09f BB |
1533 | /* If the remaining range is past end of file, we're done */ |
1534 | if ((off >> blkshift) > dn->dn_maxblkid) | |
1535 | goto out; | |
34dc7c2f | 1536 | |
b128c09f BB |
1537 | ASSERT(ISP2(blksz)); |
1538 | if (trunc) | |
1539 | tail = 0; | |
1540 | else | |
1541 | tail = P2PHASE(len, blksz); | |
1542 | ||
c99c9001 | 1543 | ASSERT0(P2PHASE(off, blksz)); |
b128c09f BB |
1544 | /* zero out any partial block data at the end of the range */ |
1545 | if (tail) { | |
1546 | if (len < tail) | |
1547 | tail = len; | |
1548 | if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off+len), | |
1549 | TRUE, FTAG, &db) == 0) { | |
1550 | /* don't dirty if not on disk and not dirty */ | |
1551 | if (db->db_last_dirty || | |
1552 | (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { | |
1553 | rw_exit(&dn->dn_struct_rwlock); | |
b0bc7a84 | 1554 | dmu_buf_will_dirty(&db->db, tx); |
b128c09f BB |
1555 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); |
1556 | bzero(db->db.db_data, tail); | |
34dc7c2f | 1557 | } |
b128c09f | 1558 | dbuf_rele(db, FTAG); |
34dc7c2f | 1559 | } |
b128c09f BB |
1560 | len -= tail; |
1561 | } | |
34dc7c2f | 1562 | |
b128c09f BB |
1563 | /* If the range did not include a full block, we are done */ |
1564 | if (len == 0) | |
1565 | goto out; | |
1566 | ||
1567 | ASSERT(IS_P2ALIGNED(off, blksz)); | |
1568 | ASSERT(trunc || IS_P2ALIGNED(len, blksz)); | |
1569 | blkid = off >> blkshift; | |
1570 | nblks = len >> blkshift; | |
1571 | if (trunc) | |
1572 | nblks += 1; | |
1573 | ||
1574 | /* | |
b0bc7a84 MG |
1575 | * Dirty the first and last indirect blocks, as they (and/or their |
1576 | * parents) will need to be written out if they were only | |
1577 | * partially freed. Interior indirect blocks will be themselves freed, | |
1578 | * by free_children(), so they need not be dirtied. Note that these | |
1579 | * interior blocks have already been prefetched by dmu_tx_hold_free(). | |
b128c09f BB |
1580 | */ |
1581 | if (dn->dn_nlevels > 1) { | |
b0bc7a84 | 1582 | uint64_t first, last; |
b128c09f BB |
1583 | |
1584 | first = blkid >> epbs; | |
c65aa5b2 | 1585 | if ((db = dbuf_hold_level(dn, 1, first, FTAG))) { |
b0bc7a84 | 1586 | dmu_buf_will_dirty(&db->db, tx); |
34dc7c2f BB |
1587 | dbuf_rele(db, FTAG); |
1588 | } | |
b128c09f BB |
1589 | if (trunc) |
1590 | last = dn->dn_maxblkid >> epbs; | |
1591 | else | |
1592 | last = (blkid + nblks - 1) >> epbs; | |
1593 | if (last > first && (db = dbuf_hold_level(dn, 1, last, FTAG))) { | |
b0bc7a84 | 1594 | dmu_buf_will_dirty(&db->db, tx); |
34dc7c2f BB |
1595 | dbuf_rele(db, FTAG); |
1596 | } | |
34dc7c2f | 1597 | } |
b0bc7a84 | 1598 | |
b128c09f BB |
1599 | done: |
1600 | /* | |
1601 | * Add this range to the dnode range list. | |
1602 | * We will finish up this free operation in the syncing phase. | |
1603 | */ | |
34dc7c2f | 1604 | mutex_enter(&dn->dn_mtx); |
34dc7c2f | 1605 | { |
9bd274dd MA |
1606 | int txgoff = tx->tx_txg & TXG_MASK; |
1607 | if (dn->dn_free_ranges[txgoff] == NULL) { | |
1608 | dn->dn_free_ranges[txgoff] = | |
1609 | range_tree_create(NULL, NULL, &dn->dn_mtx); | |
1610 | } | |
1611 | range_tree_clear(dn->dn_free_ranges[txgoff], blkid, nblks); | |
1612 | range_tree_add(dn->dn_free_ranges[txgoff], blkid, nblks); | |
34dc7c2f | 1613 | } |
9bd274dd MA |
1614 | dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", |
1615 | blkid, nblks, tx->tx_txg); | |
34dc7c2f BB |
1616 | mutex_exit(&dn->dn_mtx); |
1617 | ||
b128c09f | 1618 | dbuf_free_range(dn, blkid, blkid + nblks - 1, tx); |
34dc7c2f BB |
1619 | dnode_setdirty(dn, tx); |
1620 | out: | |
b128c09f | 1621 | |
34dc7c2f BB |
1622 | rw_exit(&dn->dn_struct_rwlock); |
1623 | } | |
1624 | ||
428870ff BB |
1625 | static boolean_t |
1626 | dnode_spill_freed(dnode_t *dn) | |
1627 | { | |
1628 | int i; | |
1629 | ||
1630 | mutex_enter(&dn->dn_mtx); | |
1631 | for (i = 0; i < TXG_SIZE; i++) { | |
1632 | if (dn->dn_rm_spillblk[i] == DN_KILL_SPILLBLK) | |
1633 | break; | |
1634 | } | |
1635 | mutex_exit(&dn->dn_mtx); | |
1636 | return (i < TXG_SIZE); | |
1637 | } | |
1638 | ||
34dc7c2f BB |
1639 | /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */ |
1640 | uint64_t | |
1641 | dnode_block_freed(dnode_t *dn, uint64_t blkid) | |
1642 | { | |
34dc7c2f BB |
1643 | void *dp = spa_get_dsl(dn->dn_objset->os_spa); |
1644 | int i; | |
1645 | ||
428870ff | 1646 | if (blkid == DMU_BONUS_BLKID) |
34dc7c2f BB |
1647 | return (FALSE); |
1648 | ||
1649 | /* | |
1650 | * If we're in the process of opening the pool, dp will not be | |
1651 | * set yet, but there shouldn't be anything dirty. | |
1652 | */ | |
1653 | if (dp == NULL) | |
1654 | return (FALSE); | |
1655 | ||
1656 | if (dn->dn_free_txg) | |
1657 | return (TRUE); | |
1658 | ||
428870ff BB |
1659 | if (blkid == DMU_SPILL_BLKID) |
1660 | return (dnode_spill_freed(dn)); | |
1661 | ||
34dc7c2f BB |
1662 | mutex_enter(&dn->dn_mtx); |
1663 | for (i = 0; i < TXG_SIZE; i++) { | |
9bd274dd MA |
1664 | if (dn->dn_free_ranges[i] != NULL && |
1665 | range_tree_contains(dn->dn_free_ranges[i], blkid, 1)) | |
34dc7c2f BB |
1666 | break; |
1667 | } | |
1668 | mutex_exit(&dn->dn_mtx); | |
1669 | return (i < TXG_SIZE); | |
1670 | } | |
1671 | ||
1672 | /* call from syncing context when we actually write/free space for this dnode */ | |
1673 | void | |
1674 | dnode_diduse_space(dnode_t *dn, int64_t delta) | |
1675 | { | |
1676 | uint64_t space; | |
1677 | dprintf_dnode(dn, "dn=%p dnp=%p used=%llu delta=%lld\n", | |
1678 | dn, dn->dn_phys, | |
1679 | (u_longlong_t)dn->dn_phys->dn_used, | |
1680 | (longlong_t)delta); | |
1681 | ||
1682 | mutex_enter(&dn->dn_mtx); | |
1683 | space = DN_USED_BYTES(dn->dn_phys); | |
1684 | if (delta > 0) { | |
1685 | ASSERT3U(space + delta, >=, space); /* no overflow */ | |
1686 | } else { | |
1687 | ASSERT3U(space, >=, -delta); /* no underflow */ | |
1688 | } | |
1689 | space += delta; | |
1690 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_DNODE_BYTES) { | |
1691 | ASSERT((dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) == 0); | |
c99c9001 | 1692 | ASSERT0(P2PHASE(space, 1<<DEV_BSHIFT)); |
34dc7c2f BB |
1693 | dn->dn_phys->dn_used = space >> DEV_BSHIFT; |
1694 | } else { | |
1695 | dn->dn_phys->dn_used = space; | |
1696 | dn->dn_phys->dn_flags |= DNODE_FLAG_USED_BYTES; | |
1697 | } | |
1698 | mutex_exit(&dn->dn_mtx); | |
1699 | } | |
1700 | ||
1701 | /* | |
e8b96c60 MA |
1702 | * Call when we think we're going to write/free space in open context to track |
1703 | * the amount of memory in use by the currently open txg. | |
34dc7c2f BB |
1704 | */ |
1705 | void | |
1706 | dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx) | |
1707 | { | |
428870ff | 1708 | objset_t *os = dn->dn_objset; |
34dc7c2f | 1709 | dsl_dataset_t *ds = os->os_dsl_dataset; |
e8b96c60 | 1710 | int64_t aspace = spa_get_asize(os->os_spa, space); |
34dc7c2f | 1711 | |
e8b96c60 MA |
1712 | if (ds != NULL) { |
1713 | dsl_dir_willuse_space(ds->ds_dir, aspace, tx); | |
1714 | dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); | |
1715 | } | |
34dc7c2f | 1716 | |
e8b96c60 | 1717 | dmu_tx_willuse_space(tx, aspace); |
34dc7c2f BB |
1718 | } |
1719 | ||
45d1cae3 | 1720 | /* |
d3cc8b15 WA |
1721 | * Scans a block at the indicated "level" looking for a hole or data, |
1722 | * depending on 'flags'. | |
1723 | * | |
1724 | * If level > 0, then we are scanning an indirect block looking at its | |
1725 | * pointers. If level == 0, then we are looking at a block of dnodes. | |
1726 | * | |
1727 | * If we don't find what we are looking for in the block, we return ESRCH. | |
1728 | * Otherwise, return with *offset pointing to the beginning (if searching | |
1729 | * forwards) or end (if searching backwards) of the range covered by the | |
1730 | * block pointer we matched on (or dnode). | |
45d1cae3 BB |
1731 | * |
1732 | * The basic search algorithm used below by dnode_next_offset() is to | |
1733 | * use this function to search up the block tree (widen the search) until | |
1734 | * we find something (i.e., we don't return ESRCH) and then search back | |
1735 | * down the tree (narrow the search) until we reach our original search | |
1736 | * level. | |
1737 | */ | |
34dc7c2f | 1738 | static int |
b128c09f | 1739 | dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset, |
34dc7c2f BB |
1740 | int lvl, uint64_t blkfill, uint64_t txg) |
1741 | { | |
1742 | dmu_buf_impl_t *db = NULL; | |
1743 | void *data = NULL; | |
1744 | uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
1745 | uint64_t epb = 1ULL << epbs; | |
1746 | uint64_t minfill, maxfill; | |
b128c09f BB |
1747 | boolean_t hole; |
1748 | int i, inc, error, span; | |
34dc7c2f BB |
1749 | |
1750 | dprintf("probing object %llu offset %llx level %d of %u\n", | |
1751 | dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels); | |
1752 | ||
9babb374 | 1753 | hole = ((flags & DNODE_FIND_HOLE) != 0); |
b128c09f BB |
1754 | inc = (flags & DNODE_FIND_BACKWARDS) ? -1 : 1; |
1755 | ASSERT(txg == 0 || !hole); | |
1756 | ||
34dc7c2f BB |
1757 | if (lvl == dn->dn_phys->dn_nlevels) { |
1758 | error = 0; | |
1759 | epb = dn->dn_phys->dn_nblkptr; | |
1760 | data = dn->dn_phys->dn_blkptr; | |
1761 | } else { | |
1762 | uint64_t blkid = dbuf_whichblock(dn, *offset) >> (epbs * lvl); | |
1763 | error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FTAG, &db); | |
1764 | if (error) { | |
b128c09f BB |
1765 | if (error != ENOENT) |
1766 | return (error); | |
1767 | if (hole) | |
1768 | return (0); | |
1769 | /* | |
1770 | * This can only happen when we are searching up | |
1771 | * the block tree for data. We don't really need to | |
1772 | * adjust the offset, as we will just end up looking | |
1773 | * at the pointer to this block in its parent, and its | |
1774 | * going to be unallocated, so we will skip over it. | |
1775 | */ | |
2e528b49 | 1776 | return (SET_ERROR(ESRCH)); |
34dc7c2f BB |
1777 | } |
1778 | error = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_HAVESTRUCT); | |
1779 | if (error) { | |
1780 | dbuf_rele(db, FTAG); | |
1781 | return (error); | |
1782 | } | |
1783 | data = db->db.db_data; | |
1784 | } | |
1785 | ||
b0bc7a84 MG |
1786 | |
1787 | if (db != NULL && txg != 0 && (db->db_blkptr == NULL || | |
1788 | db->db_blkptr->blk_birth <= txg || | |
1789 | BP_IS_HOLE(db->db_blkptr))) { | |
b128c09f BB |
1790 | /* |
1791 | * This can only happen when we are searching up the tree | |
1792 | * and these conditions mean that we need to keep climbing. | |
1793 | */ | |
2e528b49 | 1794 | error = SET_ERROR(ESRCH); |
34dc7c2f BB |
1795 | } else if (lvl == 0) { |
1796 | dnode_phys_t *dnp = data; | |
1797 | span = DNODE_SHIFT; | |
1798 | ASSERT(dn->dn_type == DMU_OT_DNODE); | |
1799 | ||
b128c09f BB |
1800 | for (i = (*offset >> span) & (blkfill - 1); |
1801 | i >= 0 && i < blkfill; i += inc) { | |
9babb374 | 1802 | if ((dnp[i].dn_type == DMU_OT_NONE) == hole) |
34dc7c2f | 1803 | break; |
b128c09f | 1804 | *offset += (1ULL << span) * inc; |
34dc7c2f | 1805 | } |
b128c09f | 1806 | if (i < 0 || i == blkfill) |
2e528b49 | 1807 | error = SET_ERROR(ESRCH); |
34dc7c2f BB |
1808 | } else { |
1809 | blkptr_t *bp = data; | |
45d1cae3 | 1810 | uint64_t start = *offset; |
34dc7c2f BB |
1811 | span = (lvl - 1) * epbs + dn->dn_datablkshift; |
1812 | minfill = 0; | |
1813 | maxfill = blkfill << ((lvl - 1) * epbs); | |
1814 | ||
1815 | if (hole) | |
1816 | maxfill--; | |
1817 | else | |
1818 | minfill++; | |
1819 | ||
45d1cae3 BB |
1820 | *offset = *offset >> span; |
1821 | for (i = BF64_GET(*offset, 0, epbs); | |
b128c09f | 1822 | i >= 0 && i < epb; i += inc) { |
9b67f605 MA |
1823 | if (BP_GET_FILL(&bp[i]) >= minfill && |
1824 | BP_GET_FILL(&bp[i]) <= maxfill && | |
b128c09f | 1825 | (hole || bp[i].blk_birth > txg)) |
34dc7c2f | 1826 | break; |
45d1cae3 BB |
1827 | if (inc > 0 || *offset > 0) |
1828 | *offset += inc; | |
1829 | } | |
1830 | *offset = *offset << span; | |
1831 | if (inc < 0) { | |
1832 | /* traversing backwards; position offset at the end */ | |
1833 | ASSERT3U(*offset, <=, start); | |
1834 | *offset = MIN(*offset + (1ULL << span) - 1, start); | |
1835 | } else if (*offset < start) { | |
1836 | *offset = start; | |
34dc7c2f | 1837 | } |
45d1cae3 | 1838 | if (i < 0 || i >= epb) |
2e528b49 | 1839 | error = SET_ERROR(ESRCH); |
34dc7c2f BB |
1840 | } |
1841 | ||
1842 | if (db) | |
1843 | dbuf_rele(db, FTAG); | |
1844 | ||
1845 | return (error); | |
1846 | } | |
1847 | ||
1848 | /* | |
1849 | * Find the next hole, data, or sparse region at or after *offset. | |
1850 | * The value 'blkfill' tells us how many items we expect to find | |
1851 | * in an L0 data block; this value is 1 for normal objects, | |
1852 | * DNODES_PER_BLOCK for the meta dnode, and some fraction of | |
1853 | * DNODES_PER_BLOCK when searching for sparse regions thereof. | |
1854 | * | |
1855 | * Examples: | |
1856 | * | |
b128c09f BB |
1857 | * dnode_next_offset(dn, flags, offset, 1, 1, 0); |
1858 | * Finds the next/previous hole/data in a file. | |
34dc7c2f BB |
1859 | * Used in dmu_offset_next(). |
1860 | * | |
b128c09f | 1861 | * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg); |
34dc7c2f BB |
1862 | * Finds the next free/allocated dnode an objset's meta-dnode. |
1863 | * Only finds objects that have new contents since txg (ie. | |
1864 | * bonus buffer changes and content removal are ignored). | |
1865 | * Used in dmu_object_next(). | |
1866 | * | |
b128c09f | 1867 | * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0); |
34dc7c2f BB |
1868 | * Finds the next L2 meta-dnode bp that's at most 1/4 full. |
1869 | * Used in dmu_object_alloc(). | |
1870 | */ | |
1871 | int | |
b128c09f | 1872 | dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset, |
34dc7c2f BB |
1873 | int minlvl, uint64_t blkfill, uint64_t txg) |
1874 | { | |
b128c09f | 1875 | uint64_t initial_offset = *offset; |
34dc7c2f BB |
1876 | int lvl, maxlvl; |
1877 | int error = 0; | |
34dc7c2f | 1878 | |
b128c09f BB |
1879 | if (!(flags & DNODE_FIND_HAVELOCK)) |
1880 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
34dc7c2f BB |
1881 | |
1882 | if (dn->dn_phys->dn_nlevels == 0) { | |
2e528b49 | 1883 | error = SET_ERROR(ESRCH); |
b128c09f | 1884 | goto out; |
34dc7c2f BB |
1885 | } |
1886 | ||
1887 | if (dn->dn_datablkshift == 0) { | |
1888 | if (*offset < dn->dn_datablksz) { | |
b128c09f | 1889 | if (flags & DNODE_FIND_HOLE) |
34dc7c2f BB |
1890 | *offset = dn->dn_datablksz; |
1891 | } else { | |
2e528b49 | 1892 | error = SET_ERROR(ESRCH); |
34dc7c2f | 1893 | } |
b128c09f | 1894 | goto out; |
34dc7c2f BB |
1895 | } |
1896 | ||
1897 | maxlvl = dn->dn_phys->dn_nlevels; | |
1898 | ||
1899 | for (lvl = minlvl; lvl <= maxlvl; lvl++) { | |
1900 | error = dnode_next_offset_level(dn, | |
b128c09f | 1901 | flags, offset, lvl, blkfill, txg); |
34dc7c2f BB |
1902 | if (error != ESRCH) |
1903 | break; | |
1904 | } | |
1905 | ||
b128c09f | 1906 | while (error == 0 && --lvl >= minlvl) { |
34dc7c2f | 1907 | error = dnode_next_offset_level(dn, |
b128c09f | 1908 | flags, offset, lvl, blkfill, txg); |
34dc7c2f BB |
1909 | } |
1910 | ||
d97aa48f MA |
1911 | /* |
1912 | * There's always a "virtual hole" at the end of the object, even | |
1913 | * if all BP's which physically exist are non-holes. | |
1914 | */ | |
1915 | if ((flags & DNODE_FIND_HOLE) && error == ESRCH && txg == 0 && | |
1916 | minlvl == 1 && blkfill == 1 && !(flags & DNODE_FIND_BACKWARDS)) { | |
1917 | error = 0; | |
1918 | } | |
1919 | ||
b128c09f BB |
1920 | if (error == 0 && (flags & DNODE_FIND_BACKWARDS ? |
1921 | initial_offset < *offset : initial_offset > *offset)) | |
2e528b49 | 1922 | error = SET_ERROR(ESRCH); |
b128c09f BB |
1923 | out: |
1924 | if (!(flags & DNODE_FIND_HAVELOCK)) | |
1925 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
1926 | |
1927 | return (error); | |
1928 | } |