]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
34dc7c2f BB |
23 | */ |
24 | ||
34dc7c2f BB |
25 | #include <sys/zfs_context.h> |
26 | #include <sys/dbuf.h> | |
27 | #include <sys/dnode.h> | |
28 | #include <sys/dmu.h> | |
29 | #include <sys/dmu_impl.h> | |
30 | #include <sys/dmu_tx.h> | |
31 | #include <sys/dmu_objset.h> | |
32 | #include <sys/dsl_dir.h> | |
33 | #include <sys/dsl_dataset.h> | |
34 | #include <sys/spa.h> | |
35 | #include <sys/zio.h> | |
36 | #include <sys/dmu_zfetch.h> | |
37 | ||
38 | static int free_range_compar(const void *node1, const void *node2); | |
39 | ||
40 | static kmem_cache_t *dnode_cache; | |
572e2857 BB |
41 | /* |
42 | * Define DNODE_STATS to turn on statistic gathering. By default, it is only | |
43 | * turned on when DEBUG is also defined. | |
44 | */ | |
45 | #ifdef DEBUG | |
46 | #define DNODE_STATS | |
47 | #endif /* DEBUG */ | |
48 | ||
49 | #ifdef DNODE_STATS | |
50 | #define DNODE_STAT_ADD(stat) ((stat)++) | |
51 | #else | |
52 | #define DNODE_STAT_ADD(stat) /* nothing */ | |
53 | #endif /* DNODE_STATS */ | |
34dc7c2f | 54 | |
1fde1e37 | 55 | ASSERTV(static dnode_phys_t dnode_phys_zero); |
34dc7c2f BB |
56 | |
57 | int zfs_default_bs = SPA_MINBLOCKSHIFT; | |
58 | int zfs_default_ibs = DN_MAX_INDBLKSHIFT; | |
59 | ||
5ac1241a | 60 | #ifdef _KERNEL |
572e2857 | 61 | static kmem_cbrc_t dnode_move(void *, void *, size_t, void *); |
5ac1241a | 62 | #endif /* _KERNEL */ |
572e2857 | 63 | |
34dc7c2f BB |
64 | /* ARGSUSED */ |
65 | static int | |
66 | dnode_cons(void *arg, void *unused, int kmflag) | |
67 | { | |
34dc7c2f | 68 | dnode_t *dn = arg; |
572e2857 | 69 | int i; |
34dc7c2f BB |
70 | |
71 | rw_init(&dn->dn_struct_rwlock, NULL, RW_DEFAULT, NULL); | |
72 | mutex_init(&dn->dn_mtx, NULL, MUTEX_DEFAULT, NULL); | |
73 | mutex_init(&dn->dn_dbufs_mtx, NULL, MUTEX_DEFAULT, NULL); | |
fb5f0bc8 BB |
74 | cv_init(&dn->dn_notxholds, NULL, CV_DEFAULT, NULL); |
75 | ||
34dc7c2f BB |
76 | refcount_create(&dn->dn_holds); |
77 | refcount_create(&dn->dn_tx_holds); | |
572e2857 BB |
78 | list_link_init(&dn->dn_link); |
79 | ||
80 | bzero(&dn->dn_next_nblkptr[0], sizeof (dn->dn_next_nblkptr)); | |
81 | bzero(&dn->dn_next_nlevels[0], sizeof (dn->dn_next_nlevels)); | |
82 | bzero(&dn->dn_next_indblkshift[0], sizeof (dn->dn_next_indblkshift)); | |
83 | bzero(&dn->dn_next_bonustype[0], sizeof (dn->dn_next_bonustype)); | |
84 | bzero(&dn->dn_rm_spillblk[0], sizeof (dn->dn_rm_spillblk)); | |
85 | bzero(&dn->dn_next_bonuslen[0], sizeof (dn->dn_next_bonuslen)); | |
86 | bzero(&dn->dn_next_blksz[0], sizeof (dn->dn_next_blksz)); | |
34dc7c2f BB |
87 | |
88 | for (i = 0; i < TXG_SIZE; i++) { | |
572e2857 | 89 | list_link_init(&dn->dn_dirty_link[i]); |
34dc7c2f BB |
90 | avl_create(&dn->dn_ranges[i], free_range_compar, |
91 | sizeof (free_range_t), | |
92 | offsetof(struct free_range, fr_node)); | |
93 | list_create(&dn->dn_dirty_records[i], | |
94 | sizeof (dbuf_dirty_record_t), | |
95 | offsetof(dbuf_dirty_record_t, dr_dirty_node)); | |
96 | } | |
97 | ||
572e2857 BB |
98 | dn->dn_allocated_txg = 0; |
99 | dn->dn_free_txg = 0; | |
100 | dn->dn_assigned_txg = 0; | |
101 | dn->dn_dirtyctx = 0; | |
102 | dn->dn_dirtyctx_firstset = NULL; | |
103 | dn->dn_bonus = NULL; | |
104 | dn->dn_have_spill = B_FALSE; | |
105 | dn->dn_zio = NULL; | |
106 | dn->dn_oldused = 0; | |
107 | dn->dn_oldflags = 0; | |
108 | dn->dn_olduid = 0; | |
109 | dn->dn_oldgid = 0; | |
110 | dn->dn_newuid = 0; | |
111 | dn->dn_newgid = 0; | |
112 | dn->dn_id_flags = 0; | |
113 | ||
114 | dn->dn_dbufs_count = 0; | |
34dc7c2f BB |
115 | list_create(&dn->dn_dbufs, sizeof (dmu_buf_impl_t), |
116 | offsetof(dmu_buf_impl_t, db_link)); | |
117 | ||
572e2857 | 118 | dn->dn_moved = 0; |
34dc7c2f BB |
119 | return (0); |
120 | } | |
121 | ||
122 | /* ARGSUSED */ | |
123 | static void | |
124 | dnode_dest(void *arg, void *unused) | |
125 | { | |
126 | int i; | |
127 | dnode_t *dn = arg; | |
128 | ||
129 | rw_destroy(&dn->dn_struct_rwlock); | |
130 | mutex_destroy(&dn->dn_mtx); | |
131 | mutex_destroy(&dn->dn_dbufs_mtx); | |
fb5f0bc8 | 132 | cv_destroy(&dn->dn_notxholds); |
34dc7c2f BB |
133 | refcount_destroy(&dn->dn_holds); |
134 | refcount_destroy(&dn->dn_tx_holds); | |
572e2857 | 135 | ASSERT(!list_link_active(&dn->dn_link)); |
34dc7c2f BB |
136 | |
137 | for (i = 0; i < TXG_SIZE; i++) { | |
572e2857 | 138 | ASSERT(!list_link_active(&dn->dn_dirty_link[i])); |
34dc7c2f BB |
139 | avl_destroy(&dn->dn_ranges[i]); |
140 | list_destroy(&dn->dn_dirty_records[i]); | |
572e2857 BB |
141 | ASSERT3U(dn->dn_next_nblkptr[i], ==, 0); |
142 | ASSERT3U(dn->dn_next_nlevels[i], ==, 0); | |
143 | ASSERT3U(dn->dn_next_indblkshift[i], ==, 0); | |
144 | ASSERT3U(dn->dn_next_bonustype[i], ==, 0); | |
145 | ASSERT3U(dn->dn_rm_spillblk[i], ==, 0); | |
146 | ASSERT3U(dn->dn_next_bonuslen[i], ==, 0); | |
147 | ASSERT3U(dn->dn_next_blksz[i], ==, 0); | |
34dc7c2f BB |
148 | } |
149 | ||
572e2857 BB |
150 | ASSERT3U(dn->dn_allocated_txg, ==, 0); |
151 | ASSERT3U(dn->dn_free_txg, ==, 0); | |
152 | ASSERT3U(dn->dn_assigned_txg, ==, 0); | |
153 | ASSERT3U(dn->dn_dirtyctx, ==, 0); | |
154 | ASSERT3P(dn->dn_dirtyctx_firstset, ==, NULL); | |
155 | ASSERT3P(dn->dn_bonus, ==, NULL); | |
156 | ASSERT(!dn->dn_have_spill); | |
157 | ASSERT3P(dn->dn_zio, ==, NULL); | |
158 | ASSERT3U(dn->dn_oldused, ==, 0); | |
159 | ASSERT3U(dn->dn_oldflags, ==, 0); | |
160 | ASSERT3U(dn->dn_olduid, ==, 0); | |
161 | ASSERT3U(dn->dn_oldgid, ==, 0); | |
162 | ASSERT3U(dn->dn_newuid, ==, 0); | |
163 | ASSERT3U(dn->dn_newgid, ==, 0); | |
164 | ASSERT3U(dn->dn_id_flags, ==, 0); | |
165 | ||
166 | ASSERT3U(dn->dn_dbufs_count, ==, 0); | |
34dc7c2f BB |
167 | list_destroy(&dn->dn_dbufs); |
168 | } | |
169 | ||
170 | void | |
171 | dnode_init(void) | |
172 | { | |
572e2857 | 173 | ASSERT(dnode_cache == NULL); |
ae6ba3db BB |
174 | dnode_cache = kmem_cache_create("dnode_t", sizeof (dnode_t), |
175 | 0, dnode_cons, dnode_dest, NULL, NULL, NULL, KMC_KMEM); | |
572e2857 | 176 | kmem_cache_set_move(dnode_cache, dnode_move); |
34dc7c2f BB |
177 | } |
178 | ||
179 | void | |
180 | dnode_fini(void) | |
181 | { | |
182 | kmem_cache_destroy(dnode_cache); | |
572e2857 | 183 | dnode_cache = NULL; |
34dc7c2f BB |
184 | } |
185 | ||
186 | ||
187 | #ifdef ZFS_DEBUG | |
188 | void | |
189 | dnode_verify(dnode_t *dn) | |
190 | { | |
191 | int drop_struct_lock = FALSE; | |
192 | ||
193 | ASSERT(dn->dn_phys); | |
194 | ASSERT(dn->dn_objset); | |
572e2857 | 195 | ASSERT(dn->dn_handle->dnh_dnode == dn); |
34dc7c2f BB |
196 | |
197 | ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); | |
198 | ||
199 | if (!(zfs_flags & ZFS_DEBUG_DNODE_VERIFY)) | |
200 | return; | |
201 | ||
202 | if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { | |
203 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
204 | drop_struct_lock = TRUE; | |
205 | } | |
206 | if (dn->dn_phys->dn_type != DMU_OT_NONE || dn->dn_allocated_txg != 0) { | |
207 | int i; | |
34dc7c2f BB |
208 | ASSERT3U(dn->dn_indblkshift, <=, SPA_MAXBLOCKSHIFT); |
209 | if (dn->dn_datablkshift) { | |
210 | ASSERT3U(dn->dn_datablkshift, >=, SPA_MINBLOCKSHIFT); | |
211 | ASSERT3U(dn->dn_datablkshift, <=, SPA_MAXBLOCKSHIFT); | |
212 | ASSERT3U(1<<dn->dn_datablkshift, ==, dn->dn_datablksz); | |
213 | } | |
214 | ASSERT3U(dn->dn_nlevels, <=, 30); | |
215 | ASSERT3U(dn->dn_type, <=, DMU_OT_NUMTYPES); | |
216 | ASSERT3U(dn->dn_nblkptr, >=, 1); | |
217 | ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); | |
218 | ASSERT3U(dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); | |
219 | ASSERT3U(dn->dn_datablksz, ==, | |
220 | dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); | |
221 | ASSERT3U(ISP2(dn->dn_datablksz), ==, dn->dn_datablkshift != 0); | |
222 | ASSERT3U((dn->dn_nblkptr - 1) * sizeof (blkptr_t) + | |
223 | dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); | |
224 | for (i = 0; i < TXG_SIZE; i++) { | |
225 | ASSERT3U(dn->dn_next_nlevels[i], <=, dn->dn_nlevels); | |
226 | } | |
227 | } | |
228 | if (dn->dn_phys->dn_type != DMU_OT_NONE) | |
229 | ASSERT3U(dn->dn_phys->dn_nlevels, <=, dn->dn_nlevels); | |
9babb374 | 230 | ASSERT(DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_dbuf != NULL); |
34dc7c2f BB |
231 | if (dn->dn_dbuf != NULL) { |
232 | ASSERT3P(dn->dn_phys, ==, | |
233 | (dnode_phys_t *)dn->dn_dbuf->db.db_data + | |
234 | (dn->dn_object % (dn->dn_dbuf->db.db_size >> DNODE_SHIFT))); | |
235 | } | |
236 | if (drop_struct_lock) | |
237 | rw_exit(&dn->dn_struct_rwlock); | |
238 | } | |
239 | #endif | |
240 | ||
241 | void | |
242 | dnode_byteswap(dnode_phys_t *dnp) | |
243 | { | |
244 | uint64_t *buf64 = (void*)&dnp->dn_blkptr; | |
245 | int i; | |
246 | ||
247 | if (dnp->dn_type == DMU_OT_NONE) { | |
248 | bzero(dnp, sizeof (dnode_phys_t)); | |
249 | return; | |
250 | } | |
251 | ||
252 | dnp->dn_datablkszsec = BSWAP_16(dnp->dn_datablkszsec); | |
253 | dnp->dn_bonuslen = BSWAP_16(dnp->dn_bonuslen); | |
254 | dnp->dn_maxblkid = BSWAP_64(dnp->dn_maxblkid); | |
255 | dnp->dn_used = BSWAP_64(dnp->dn_used); | |
256 | ||
257 | /* | |
258 | * dn_nblkptr is only one byte, so it's OK to read it in either | |
259 | * byte order. We can't read dn_bouslen. | |
260 | */ | |
261 | ASSERT(dnp->dn_indblkshift <= SPA_MAXBLOCKSHIFT); | |
262 | ASSERT(dnp->dn_nblkptr <= DN_MAX_NBLKPTR); | |
263 | for (i = 0; i < dnp->dn_nblkptr * sizeof (blkptr_t)/8; i++) | |
264 | buf64[i] = BSWAP_64(buf64[i]); | |
265 | ||
266 | /* | |
267 | * OK to check dn_bonuslen for zero, because it won't matter if | |
268 | * we have the wrong byte order. This is necessary because the | |
269 | * dnode dnode is smaller than a regular dnode. | |
270 | */ | |
271 | if (dnp->dn_bonuslen != 0) { | |
272 | /* | |
273 | * Note that the bonus length calculated here may be | |
274 | * longer than the actual bonus buffer. This is because | |
275 | * we always put the bonus buffer after the last block | |
276 | * pointer (instead of packing it against the end of the | |
277 | * dnode buffer). | |
278 | */ | |
279 | int off = (dnp->dn_nblkptr-1) * sizeof (blkptr_t); | |
280 | size_t len = DN_MAX_BONUSLEN - off; | |
281 | ASSERT3U(dnp->dn_bonustype, <, DMU_OT_NUMTYPES); | |
282 | dmu_ot[dnp->dn_bonustype].ot_byteswap(dnp->dn_bonus + off, len); | |
283 | } | |
428870ff BB |
284 | |
285 | /* Swap SPILL block if we have one */ | |
286 | if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) | |
287 | byteswap_uint64_array(&dnp->dn_spill, sizeof (blkptr_t)); | |
288 | ||
34dc7c2f BB |
289 | } |
290 | ||
291 | void | |
292 | dnode_buf_byteswap(void *vbuf, size_t size) | |
293 | { | |
294 | dnode_phys_t *buf = vbuf; | |
295 | int i; | |
296 | ||
297 | ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT)); | |
298 | ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0); | |
299 | ||
300 | size >>= DNODE_SHIFT; | |
301 | for (i = 0; i < size; i++) { | |
302 | dnode_byteswap(buf); | |
303 | buf++; | |
304 | } | |
305 | } | |
306 | ||
307 | static int | |
308 | free_range_compar(const void *node1, const void *node2) | |
309 | { | |
310 | const free_range_t *rp1 = node1; | |
311 | const free_range_t *rp2 = node2; | |
312 | ||
313 | if (rp1->fr_blkid < rp2->fr_blkid) | |
314 | return (-1); | |
315 | else if (rp1->fr_blkid > rp2->fr_blkid) | |
316 | return (1); | |
317 | else return (0); | |
318 | } | |
319 | ||
320 | void | |
321 | dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx) | |
322 | { | |
323 | ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); | |
324 | ||
325 | dnode_setdirty(dn, tx); | |
326 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
327 | ASSERT3U(newsize, <=, DN_MAX_BONUSLEN - | |
328 | (dn->dn_nblkptr-1) * sizeof (blkptr_t)); | |
329 | dn->dn_bonuslen = newsize; | |
330 | if (newsize == 0) | |
331 | dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = DN_ZERO_BONUSLEN; | |
332 | else | |
333 | dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; | |
334 | rw_exit(&dn->dn_struct_rwlock); | |
335 | } | |
336 | ||
428870ff BB |
337 | void |
338 | dnode_setbonus_type(dnode_t *dn, dmu_object_type_t newtype, dmu_tx_t *tx) | |
339 | { | |
340 | ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); | |
341 | dnode_setdirty(dn, tx); | |
342 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
343 | dn->dn_bonustype = newtype; | |
344 | dn->dn_next_bonustype[tx->tx_txg & TXG_MASK] = dn->dn_bonustype; | |
345 | rw_exit(&dn->dn_struct_rwlock); | |
346 | } | |
347 | ||
348 | void | |
349 | dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx) | |
350 | { | |
351 | ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); | |
352 | ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); | |
353 | dnode_setdirty(dn, tx); | |
354 | dn->dn_rm_spillblk[tx->tx_txg&TXG_MASK] = DN_KILL_SPILLBLK; | |
355 | dn->dn_have_spill = B_FALSE; | |
356 | } | |
357 | ||
34dc7c2f BB |
358 | static void |
359 | dnode_setdblksz(dnode_t *dn, int size) | |
360 | { | |
361 | ASSERT3U(P2PHASE(size, SPA_MINBLOCKSIZE), ==, 0); | |
362 | ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); | |
363 | ASSERT3U(size, >=, SPA_MINBLOCKSIZE); | |
364 | ASSERT3U(size >> SPA_MINBLOCKSHIFT, <, | |
365 | 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8)); | |
366 | dn->dn_datablksz = size; | |
367 | dn->dn_datablkszsec = size >> SPA_MINBLOCKSHIFT; | |
368 | dn->dn_datablkshift = ISP2(size) ? highbit(size - 1) : 0; | |
369 | } | |
370 | ||
371 | static dnode_t * | |
428870ff | 372 | dnode_create(objset_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, |
572e2857 | 373 | uint64_t object, dnode_handle_t *dnh) |
34dc7c2f | 374 | { |
b8d06fca | 375 | dnode_t *dn = kmem_cache_alloc(dnode_cache, KM_PUSHPAGE); |
34dc7c2f | 376 | |
572e2857 BB |
377 | ASSERT(!POINTER_IS_VALID(dn->dn_objset)); |
378 | dn->dn_moved = 0; | |
379 | ||
380 | /* | |
381 | * Defer setting dn_objset until the dnode is ready to be a candidate | |
382 | * for the dnode_move() callback. | |
383 | */ | |
34dc7c2f BB |
384 | dn->dn_object = object; |
385 | dn->dn_dbuf = db; | |
572e2857 | 386 | dn->dn_handle = dnh; |
34dc7c2f BB |
387 | dn->dn_phys = dnp; |
388 | ||
572e2857 | 389 | if (dnp->dn_datablkszsec) { |
34dc7c2f | 390 | dnode_setdblksz(dn, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); |
572e2857 BB |
391 | } else { |
392 | dn->dn_datablksz = 0; | |
393 | dn->dn_datablkszsec = 0; | |
394 | dn->dn_datablkshift = 0; | |
395 | } | |
34dc7c2f BB |
396 | dn->dn_indblkshift = dnp->dn_indblkshift; |
397 | dn->dn_nlevels = dnp->dn_nlevels; | |
398 | dn->dn_type = dnp->dn_type; | |
399 | dn->dn_nblkptr = dnp->dn_nblkptr; | |
400 | dn->dn_checksum = dnp->dn_checksum; | |
401 | dn->dn_compress = dnp->dn_compress; | |
402 | dn->dn_bonustype = dnp->dn_bonustype; | |
403 | dn->dn_bonuslen = dnp->dn_bonuslen; | |
404 | dn->dn_maxblkid = dnp->dn_maxblkid; | |
428870ff BB |
405 | dn->dn_have_spill = ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0); |
406 | dn->dn_id_flags = 0; | |
34dc7c2f BB |
407 | |
408 | dmu_zfetch_init(&dn->dn_zfetch, dn); | |
409 | ||
410 | ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES); | |
572e2857 | 411 | |
34dc7c2f BB |
412 | mutex_enter(&os->os_lock); |
413 | list_insert_head(&os->os_dnodes, dn); | |
572e2857 BB |
414 | membar_producer(); |
415 | /* | |
416 | * Everything else must be valid before assigning dn_objset makes the | |
417 | * dnode eligible for dnode_move(). | |
418 | */ | |
419 | dn->dn_objset = os; | |
34dc7c2f BB |
420 | mutex_exit(&os->os_lock); |
421 | ||
d164b209 | 422 | arc_space_consume(sizeof (dnode_t), ARC_SPACE_OTHER); |
34dc7c2f BB |
423 | return (dn); |
424 | } | |
425 | ||
572e2857 BB |
426 | /* |
427 | * Caller must be holding the dnode handle, which is released upon return. | |
428 | */ | |
34dc7c2f BB |
429 | static void |
430 | dnode_destroy(dnode_t *dn) | |
431 | { | |
428870ff | 432 | objset_t *os = dn->dn_objset; |
34dc7c2f | 433 | |
428870ff | 434 | ASSERT((dn->dn_id_flags & DN_ID_NEW_EXIST) == 0); |
34dc7c2f BB |
435 | |
436 | mutex_enter(&os->os_lock); | |
572e2857 | 437 | POINTER_INVALIDATE(&dn->dn_objset); |
34dc7c2f BB |
438 | list_remove(&os->os_dnodes, dn); |
439 | mutex_exit(&os->os_lock); | |
440 | ||
572e2857 BB |
441 | /* the dnode can no longer move, so we can release the handle */ |
442 | zrl_remove(&dn->dn_handle->dnh_zrlock); | |
443 | ||
444 | dn->dn_allocated_txg = 0; | |
445 | dn->dn_free_txg = 0; | |
446 | dn->dn_assigned_txg = 0; | |
447 | ||
448 | dn->dn_dirtyctx = 0; | |
449 | if (dn->dn_dirtyctx_firstset != NULL) { | |
34dc7c2f BB |
450 | kmem_free(dn->dn_dirtyctx_firstset, 1); |
451 | dn->dn_dirtyctx_firstset = NULL; | |
452 | } | |
572e2857 | 453 | if (dn->dn_bonus != NULL) { |
34dc7c2f BB |
454 | mutex_enter(&dn->dn_bonus->db_mtx); |
455 | dbuf_evict(dn->dn_bonus); | |
456 | dn->dn_bonus = NULL; | |
457 | } | |
572e2857 BB |
458 | dn->dn_zio = NULL; |
459 | ||
460 | dn->dn_have_spill = B_FALSE; | |
461 | dn->dn_oldused = 0; | |
462 | dn->dn_oldflags = 0; | |
463 | dn->dn_olduid = 0; | |
464 | dn->dn_oldgid = 0; | |
465 | dn->dn_newuid = 0; | |
466 | dn->dn_newgid = 0; | |
467 | dn->dn_id_flags = 0; | |
468 | ||
469 | dmu_zfetch_rele(&dn->dn_zfetch); | |
34dc7c2f | 470 | kmem_cache_free(dnode_cache, dn); |
d164b209 | 471 | arc_space_return(sizeof (dnode_t), ARC_SPACE_OTHER); |
34dc7c2f BB |
472 | } |
473 | ||
474 | void | |
475 | dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, | |
476 | dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) | |
477 | { | |
478 | int i; | |
479 | ||
480 | if (blocksize == 0) | |
481 | blocksize = 1 << zfs_default_bs; | |
482 | else if (blocksize > SPA_MAXBLOCKSIZE) | |
483 | blocksize = SPA_MAXBLOCKSIZE; | |
484 | else | |
485 | blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); | |
486 | ||
487 | if (ibs == 0) | |
488 | ibs = zfs_default_ibs; | |
489 | ||
490 | ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT); | |
491 | ||
492 | dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn->dn_objset, | |
493 | dn->dn_object, tx->tx_txg, blocksize, ibs); | |
494 | ||
495 | ASSERT(dn->dn_type == DMU_OT_NONE); | |
496 | ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); | |
497 | ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); | |
498 | ASSERT(ot != DMU_OT_NONE); | |
499 | ASSERT3U(ot, <, DMU_OT_NUMTYPES); | |
500 | ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || | |
428870ff | 501 | (bonustype == DMU_OT_SA && bonuslen == 0) || |
34dc7c2f BB |
502 | (bonustype != DMU_OT_NONE && bonuslen != 0)); |
503 | ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); | |
504 | ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); | |
505 | ASSERT(dn->dn_type == DMU_OT_NONE); | |
506 | ASSERT3U(dn->dn_maxblkid, ==, 0); | |
507 | ASSERT3U(dn->dn_allocated_txg, ==, 0); | |
508 | ASSERT3U(dn->dn_assigned_txg, ==, 0); | |
509 | ASSERT(refcount_is_zero(&dn->dn_tx_holds)); | |
510 | ASSERT3U(refcount_count(&dn->dn_holds), <=, 1); | |
511 | ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL); | |
512 | ||
513 | for (i = 0; i < TXG_SIZE; i++) { | |
572e2857 | 514 | ASSERT3U(dn->dn_next_nblkptr[i], ==, 0); |
34dc7c2f BB |
515 | ASSERT3U(dn->dn_next_nlevels[i], ==, 0); |
516 | ASSERT3U(dn->dn_next_indblkshift[i], ==, 0); | |
517 | ASSERT3U(dn->dn_next_bonuslen[i], ==, 0); | |
428870ff BB |
518 | ASSERT3U(dn->dn_next_bonustype[i], ==, 0); |
519 | ASSERT3U(dn->dn_rm_spillblk[i], ==, 0); | |
34dc7c2f BB |
520 | ASSERT3U(dn->dn_next_blksz[i], ==, 0); |
521 | ASSERT(!list_link_active(&dn->dn_dirty_link[i])); | |
522 | ASSERT3P(list_head(&dn->dn_dirty_records[i]), ==, NULL); | |
523 | ASSERT3U(avl_numnodes(&dn->dn_ranges[i]), ==, 0); | |
524 | } | |
525 | ||
526 | dn->dn_type = ot; | |
527 | dnode_setdblksz(dn, blocksize); | |
528 | dn->dn_indblkshift = ibs; | |
529 | dn->dn_nlevels = 1; | |
428870ff BB |
530 | if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */ |
531 | dn->dn_nblkptr = 1; | |
532 | else | |
533 | dn->dn_nblkptr = 1 + | |
534 | ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); | |
34dc7c2f BB |
535 | dn->dn_bonustype = bonustype; |
536 | dn->dn_bonuslen = bonuslen; | |
537 | dn->dn_checksum = ZIO_CHECKSUM_INHERIT; | |
538 | dn->dn_compress = ZIO_COMPRESS_INHERIT; | |
539 | dn->dn_dirtyctx = 0; | |
540 | ||
541 | dn->dn_free_txg = 0; | |
542 | if (dn->dn_dirtyctx_firstset) { | |
543 | kmem_free(dn->dn_dirtyctx_firstset, 1); | |
544 | dn->dn_dirtyctx_firstset = NULL; | |
545 | } | |
546 | ||
547 | dn->dn_allocated_txg = tx->tx_txg; | |
428870ff | 548 | dn->dn_id_flags = 0; |
34dc7c2f BB |
549 | |
550 | dnode_setdirty(dn, tx); | |
551 | dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs; | |
552 | dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; | |
428870ff | 553 | dn->dn_next_bonustype[tx->tx_txg & TXG_MASK] = dn->dn_bonustype; |
34dc7c2f BB |
554 | dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz; |
555 | } | |
556 | ||
557 | void | |
558 | dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, | |
559 | dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) | |
560 | { | |
9babb374 | 561 | int nblkptr; |
34dc7c2f BB |
562 | |
563 | ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); | |
564 | ASSERT3U(blocksize, <=, SPA_MAXBLOCKSIZE); | |
565 | ASSERT3U(blocksize % SPA_MINBLOCKSIZE, ==, 0); | |
566 | ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); | |
567 | ASSERT(tx->tx_txg != 0); | |
568 | ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || | |
428870ff BB |
569 | (bonustype != DMU_OT_NONE && bonuslen != 0) || |
570 | (bonustype == DMU_OT_SA && bonuslen == 0)); | |
34dc7c2f BB |
571 | ASSERT3U(bonustype, <, DMU_OT_NUMTYPES); |
572 | ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); | |
573 | ||
34dc7c2f BB |
574 | /* clean up any unreferenced dbufs */ |
575 | dnode_evict_dbufs(dn); | |
d164b209 | 576 | |
428870ff BB |
577 | dn->dn_id_flags = 0; |
578 | ||
34dc7c2f | 579 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); |
34dc7c2f | 580 | dnode_setdirty(dn, tx); |
9babb374 BB |
581 | if (dn->dn_datablksz != blocksize) { |
582 | /* change blocksize */ | |
583 | ASSERT(dn->dn_maxblkid == 0 && | |
584 | (BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || | |
585 | dnode_block_freed(dn, 0))); | |
586 | dnode_setdblksz(dn, blocksize); | |
587 | dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize; | |
588 | } | |
589 | if (dn->dn_bonuslen != bonuslen) | |
590 | dn->dn_next_bonuslen[tx->tx_txg&TXG_MASK] = bonuslen; | |
428870ff BB |
591 | |
592 | if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */ | |
593 | nblkptr = 1; | |
594 | else | |
595 | nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); | |
596 | if (dn->dn_bonustype != bonustype) | |
597 | dn->dn_next_bonustype[tx->tx_txg&TXG_MASK] = bonustype; | |
d164b209 BB |
598 | if (dn->dn_nblkptr != nblkptr) |
599 | dn->dn_next_nblkptr[tx->tx_txg&TXG_MASK] = nblkptr; | |
428870ff BB |
600 | if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { |
601 | dbuf_rm_spill(dn, tx); | |
602 | dnode_rm_spill(dn, tx); | |
603 | } | |
34dc7c2f | 604 | rw_exit(&dn->dn_struct_rwlock); |
34dc7c2f BB |
605 | |
606 | /* change type */ | |
607 | dn->dn_type = ot; | |
608 | ||
609 | /* change bonus size and type */ | |
610 | mutex_enter(&dn->dn_mtx); | |
34dc7c2f BB |
611 | dn->dn_bonustype = bonustype; |
612 | dn->dn_bonuslen = bonuslen; | |
d164b209 | 613 | dn->dn_nblkptr = nblkptr; |
34dc7c2f BB |
614 | dn->dn_checksum = ZIO_CHECKSUM_INHERIT; |
615 | dn->dn_compress = ZIO_COMPRESS_INHERIT; | |
616 | ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); | |
617 | ||
d164b209 BB |
618 | /* fix up the bonus db_size */ |
619 | if (dn->dn_bonus) { | |
34dc7c2f BB |
620 | dn->dn_bonus->db.db_size = |
621 | DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t); | |
622 | ASSERT(dn->dn_bonuslen <= dn->dn_bonus->db.db_size); | |
623 | } | |
624 | ||
625 | dn->dn_allocated_txg = tx->tx_txg; | |
626 | mutex_exit(&dn->dn_mtx); | |
627 | } | |
628 | ||
5ac1241a | 629 | #ifdef _KERNEL |
572e2857 BB |
630 | #ifdef DNODE_STATS |
631 | static struct { | |
632 | uint64_t dms_dnode_invalid; | |
633 | uint64_t dms_dnode_recheck1; | |
634 | uint64_t dms_dnode_recheck2; | |
635 | uint64_t dms_dnode_special; | |
636 | uint64_t dms_dnode_handle; | |
637 | uint64_t dms_dnode_rwlock; | |
638 | uint64_t dms_dnode_active; | |
639 | } dnode_move_stats; | |
640 | #endif /* DNODE_STATS */ | |
641 | ||
642 | static void | |
643 | dnode_move_impl(dnode_t *odn, dnode_t *ndn) | |
644 | { | |
645 | int i; | |
646 | ||
647 | ASSERT(!RW_LOCK_HELD(&odn->dn_struct_rwlock)); | |
648 | ASSERT(MUTEX_NOT_HELD(&odn->dn_mtx)); | |
649 | ASSERT(MUTEX_NOT_HELD(&odn->dn_dbufs_mtx)); | |
650 | ASSERT(!RW_LOCK_HELD(&odn->dn_zfetch.zf_rwlock)); | |
651 | ||
652 | /* Copy fields. */ | |
653 | ndn->dn_objset = odn->dn_objset; | |
654 | ndn->dn_object = odn->dn_object; | |
655 | ndn->dn_dbuf = odn->dn_dbuf; | |
656 | ndn->dn_handle = odn->dn_handle; | |
657 | ndn->dn_phys = odn->dn_phys; | |
658 | ndn->dn_type = odn->dn_type; | |
659 | ndn->dn_bonuslen = odn->dn_bonuslen; | |
660 | ndn->dn_bonustype = odn->dn_bonustype; | |
661 | ndn->dn_nblkptr = odn->dn_nblkptr; | |
662 | ndn->dn_checksum = odn->dn_checksum; | |
663 | ndn->dn_compress = odn->dn_compress; | |
664 | ndn->dn_nlevels = odn->dn_nlevels; | |
665 | ndn->dn_indblkshift = odn->dn_indblkshift; | |
666 | ndn->dn_datablkshift = odn->dn_datablkshift; | |
667 | ndn->dn_datablkszsec = odn->dn_datablkszsec; | |
668 | ndn->dn_datablksz = odn->dn_datablksz; | |
669 | ndn->dn_maxblkid = odn->dn_maxblkid; | |
670 | bcopy(&odn->dn_next_nblkptr[0], &ndn->dn_next_nblkptr[0], | |
671 | sizeof (odn->dn_next_nblkptr)); | |
672 | bcopy(&odn->dn_next_nlevels[0], &ndn->dn_next_nlevels[0], | |
673 | sizeof (odn->dn_next_nlevels)); | |
674 | bcopy(&odn->dn_next_indblkshift[0], &ndn->dn_next_indblkshift[0], | |
675 | sizeof (odn->dn_next_indblkshift)); | |
676 | bcopy(&odn->dn_next_bonustype[0], &ndn->dn_next_bonustype[0], | |
677 | sizeof (odn->dn_next_bonustype)); | |
678 | bcopy(&odn->dn_rm_spillblk[0], &ndn->dn_rm_spillblk[0], | |
679 | sizeof (odn->dn_rm_spillblk)); | |
680 | bcopy(&odn->dn_next_bonuslen[0], &ndn->dn_next_bonuslen[0], | |
681 | sizeof (odn->dn_next_bonuslen)); | |
682 | bcopy(&odn->dn_next_blksz[0], &ndn->dn_next_blksz[0], | |
683 | sizeof (odn->dn_next_blksz)); | |
684 | for (i = 0; i < TXG_SIZE; i++) { | |
685 | list_move_tail(&ndn->dn_dirty_records[i], | |
686 | &odn->dn_dirty_records[i]); | |
687 | } | |
688 | bcopy(&odn->dn_ranges[0], &ndn->dn_ranges[0], sizeof (odn->dn_ranges)); | |
689 | ndn->dn_allocated_txg = odn->dn_allocated_txg; | |
690 | ndn->dn_free_txg = odn->dn_free_txg; | |
691 | ndn->dn_assigned_txg = odn->dn_assigned_txg; | |
692 | ndn->dn_dirtyctx = odn->dn_dirtyctx; | |
693 | ndn->dn_dirtyctx_firstset = odn->dn_dirtyctx_firstset; | |
694 | ASSERT(refcount_count(&odn->dn_tx_holds) == 0); | |
695 | refcount_transfer(&ndn->dn_holds, &odn->dn_holds); | |
696 | ASSERT(list_is_empty(&ndn->dn_dbufs)); | |
697 | list_move_tail(&ndn->dn_dbufs, &odn->dn_dbufs); | |
698 | ndn->dn_dbufs_count = odn->dn_dbufs_count; | |
699 | ndn->dn_bonus = odn->dn_bonus; | |
700 | ndn->dn_have_spill = odn->dn_have_spill; | |
701 | ndn->dn_zio = odn->dn_zio; | |
702 | ndn->dn_oldused = odn->dn_oldused; | |
703 | ndn->dn_oldflags = odn->dn_oldflags; | |
704 | ndn->dn_olduid = odn->dn_olduid; | |
705 | ndn->dn_oldgid = odn->dn_oldgid; | |
706 | ndn->dn_newuid = odn->dn_newuid; | |
707 | ndn->dn_newgid = odn->dn_newgid; | |
708 | ndn->dn_id_flags = odn->dn_id_flags; | |
709 | dmu_zfetch_init(&ndn->dn_zfetch, NULL); | |
710 | list_move_tail(&ndn->dn_zfetch.zf_stream, &odn->dn_zfetch.zf_stream); | |
711 | ndn->dn_zfetch.zf_dnode = odn->dn_zfetch.zf_dnode; | |
712 | ndn->dn_zfetch.zf_stream_cnt = odn->dn_zfetch.zf_stream_cnt; | |
713 | ndn->dn_zfetch.zf_alloc_fail = odn->dn_zfetch.zf_alloc_fail; | |
714 | ||
715 | /* | |
716 | * Update back pointers. Updating the handle fixes the back pointer of | |
717 | * every descendant dbuf as well as the bonus dbuf. | |
718 | */ | |
719 | ASSERT(ndn->dn_handle->dnh_dnode == odn); | |
720 | ndn->dn_handle->dnh_dnode = ndn; | |
721 | if (ndn->dn_zfetch.zf_dnode == odn) { | |
722 | ndn->dn_zfetch.zf_dnode = ndn; | |
723 | } | |
724 | ||
725 | /* | |
726 | * Invalidate the original dnode by clearing all of its back pointers. | |
727 | */ | |
728 | odn->dn_dbuf = NULL; | |
729 | odn->dn_handle = NULL; | |
730 | list_create(&odn->dn_dbufs, sizeof (dmu_buf_impl_t), | |
731 | offsetof(dmu_buf_impl_t, db_link)); | |
732 | odn->dn_dbufs_count = 0; | |
733 | odn->dn_bonus = NULL; | |
734 | odn->dn_zfetch.zf_dnode = NULL; | |
735 | ||
736 | /* | |
737 | * Set the low bit of the objset pointer to ensure that dnode_move() | |
738 | * recognizes the dnode as invalid in any subsequent callback. | |
739 | */ | |
740 | POINTER_INVALIDATE(&odn->dn_objset); | |
741 | ||
742 | /* | |
743 | * Satisfy the destructor. | |
744 | */ | |
745 | for (i = 0; i < TXG_SIZE; i++) { | |
746 | list_create(&odn->dn_dirty_records[i], | |
747 | sizeof (dbuf_dirty_record_t), | |
748 | offsetof(dbuf_dirty_record_t, dr_dirty_node)); | |
749 | odn->dn_ranges[i].avl_root = NULL; | |
750 | odn->dn_ranges[i].avl_numnodes = 0; | |
751 | odn->dn_next_nlevels[i] = 0; | |
752 | odn->dn_next_indblkshift[i] = 0; | |
753 | odn->dn_next_bonustype[i] = 0; | |
754 | odn->dn_rm_spillblk[i] = 0; | |
755 | odn->dn_next_bonuslen[i] = 0; | |
756 | odn->dn_next_blksz[i] = 0; | |
757 | } | |
758 | odn->dn_allocated_txg = 0; | |
759 | odn->dn_free_txg = 0; | |
760 | odn->dn_assigned_txg = 0; | |
761 | odn->dn_dirtyctx = 0; | |
762 | odn->dn_dirtyctx_firstset = NULL; | |
763 | odn->dn_have_spill = B_FALSE; | |
764 | odn->dn_zio = NULL; | |
765 | odn->dn_oldused = 0; | |
766 | odn->dn_oldflags = 0; | |
767 | odn->dn_olduid = 0; | |
768 | odn->dn_oldgid = 0; | |
769 | odn->dn_newuid = 0; | |
770 | odn->dn_newgid = 0; | |
771 | odn->dn_id_flags = 0; | |
772 | ||
773 | /* | |
774 | * Mark the dnode. | |
775 | */ | |
776 | ndn->dn_moved = 1; | |
777 | odn->dn_moved = (uint8_t)-1; | |
778 | } | |
779 | ||
572e2857 BB |
780 | /*ARGSUSED*/ |
781 | static kmem_cbrc_t | |
782 | dnode_move(void *buf, void *newbuf, size_t size, void *arg) | |
783 | { | |
784 | dnode_t *odn = buf, *ndn = newbuf; | |
785 | objset_t *os; | |
786 | int64_t refcount; | |
787 | uint32_t dbufs; | |
788 | ||
789 | /* | |
790 | * The dnode is on the objset's list of known dnodes if the objset | |
791 | * pointer is valid. We set the low bit of the objset pointer when | |
792 | * freeing the dnode to invalidate it, and the memory patterns written | |
793 | * by kmem (baddcafe and deadbeef) set at least one of the two low bits. | |
794 | * A newly created dnode sets the objset pointer last of all to indicate | |
795 | * that the dnode is known and in a valid state to be moved by this | |
796 | * function. | |
797 | */ | |
798 | os = odn->dn_objset; | |
799 | if (!POINTER_IS_VALID(os)) { | |
800 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_invalid); | |
801 | return (KMEM_CBRC_DONT_KNOW); | |
802 | } | |
803 | ||
804 | /* | |
805 | * Ensure that the objset does not go away during the move. | |
806 | */ | |
807 | rw_enter(&os_lock, RW_WRITER); | |
808 | if (os != odn->dn_objset) { | |
809 | rw_exit(&os_lock); | |
810 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_recheck1); | |
811 | return (KMEM_CBRC_DONT_KNOW); | |
812 | } | |
813 | ||
814 | /* | |
815 | * If the dnode is still valid, then so is the objset. We know that no | |
816 | * valid objset can be freed while we hold os_lock, so we can safely | |
817 | * ensure that the objset remains in use. | |
818 | */ | |
819 | mutex_enter(&os->os_lock); | |
820 | ||
821 | /* | |
822 | * Recheck the objset pointer in case the dnode was removed just before | |
823 | * acquiring the lock. | |
824 | */ | |
825 | if (os != odn->dn_objset) { | |
826 | mutex_exit(&os->os_lock); | |
827 | rw_exit(&os_lock); | |
828 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_recheck2); | |
829 | return (KMEM_CBRC_DONT_KNOW); | |
830 | } | |
831 | ||
832 | /* | |
833 | * At this point we know that as long as we hold os->os_lock, the dnode | |
834 | * cannot be freed and fields within the dnode can be safely accessed. | |
835 | * The objset listing this dnode cannot go away as long as this dnode is | |
836 | * on its list. | |
837 | */ | |
838 | rw_exit(&os_lock); | |
839 | if (DMU_OBJECT_IS_SPECIAL(odn->dn_object)) { | |
840 | mutex_exit(&os->os_lock); | |
841 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_special); | |
842 | return (KMEM_CBRC_NO); | |
843 | } | |
844 | ASSERT(odn->dn_dbuf != NULL); /* only "special" dnodes have no parent */ | |
845 | ||
846 | /* | |
847 | * Lock the dnode handle to prevent the dnode from obtaining any new | |
848 | * holds. This also prevents the descendant dbufs and the bonus dbuf | |
849 | * from accessing the dnode, so that we can discount their holds. The | |
850 | * handle is safe to access because we know that while the dnode cannot | |
851 | * go away, neither can its handle. Once we hold dnh_zrlock, we can | |
852 | * safely move any dnode referenced only by dbufs. | |
853 | */ | |
854 | if (!zrl_tryenter(&odn->dn_handle->dnh_zrlock)) { | |
855 | mutex_exit(&os->os_lock); | |
856 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_handle); | |
857 | return (KMEM_CBRC_LATER); | |
858 | } | |
859 | ||
860 | /* | |
861 | * Ensure a consistent view of the dnode's holds and the dnode's dbufs. | |
862 | * We need to guarantee that there is a hold for every dbuf in order to | |
863 | * determine whether the dnode is actively referenced. Falsely matching | |
864 | * a dbuf to an active hold would lead to an unsafe move. It's possible | |
865 | * that a thread already having an active dnode hold is about to add a | |
866 | * dbuf, and we can't compare hold and dbuf counts while the add is in | |
867 | * progress. | |
868 | */ | |
869 | if (!rw_tryenter(&odn->dn_struct_rwlock, RW_WRITER)) { | |
870 | zrl_exit(&odn->dn_handle->dnh_zrlock); | |
871 | mutex_exit(&os->os_lock); | |
872 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_rwlock); | |
873 | return (KMEM_CBRC_LATER); | |
874 | } | |
875 | ||
876 | /* | |
877 | * A dbuf may be removed (evicted) without an active dnode hold. In that | |
878 | * case, the dbuf count is decremented under the handle lock before the | |
879 | * dbuf's hold is released. This order ensures that if we count the hold | |
880 | * after the dbuf is removed but before its hold is released, we will | |
881 | * treat the unmatched hold as active and exit safely. If we count the | |
882 | * hold before the dbuf is removed, the hold is discounted, and the | |
883 | * removal is blocked until the move completes. | |
884 | */ | |
885 | refcount = refcount_count(&odn->dn_holds); | |
886 | ASSERT(refcount >= 0); | |
887 | dbufs = odn->dn_dbufs_count; | |
888 | ||
889 | /* We can't have more dbufs than dnode holds. */ | |
890 | ASSERT3U(dbufs, <=, refcount); | |
891 | DTRACE_PROBE3(dnode__move, dnode_t *, odn, int64_t, refcount, | |
892 | uint32_t, dbufs); | |
893 | ||
894 | if (refcount > dbufs) { | |
895 | rw_exit(&odn->dn_struct_rwlock); | |
896 | zrl_exit(&odn->dn_handle->dnh_zrlock); | |
897 | mutex_exit(&os->os_lock); | |
898 | DNODE_STAT_ADD(dnode_move_stats.dms_dnode_active); | |
899 | return (KMEM_CBRC_LATER); | |
900 | } | |
901 | ||
902 | rw_exit(&odn->dn_struct_rwlock); | |
903 | ||
904 | /* | |
905 | * At this point we know that anyone with a hold on the dnode is not | |
906 | * actively referencing it. The dnode is known and in a valid state to | |
907 | * move. We're holding the locks needed to execute the critical section. | |
908 | */ | |
909 | dnode_move_impl(odn, ndn); | |
910 | ||
911 | list_link_replace(&odn->dn_link, &ndn->dn_link); | |
912 | /* If the dnode was safe to move, the refcount cannot have changed. */ | |
913 | ASSERT(refcount == refcount_count(&ndn->dn_holds)); | |
914 | ASSERT(dbufs == ndn->dn_dbufs_count); | |
915 | zrl_exit(&ndn->dn_handle->dnh_zrlock); /* handle has moved */ | |
916 | mutex_exit(&os->os_lock); | |
917 | ||
918 | return (KMEM_CBRC_YES); | |
919 | } | |
920 | #endif /* _KERNEL */ | |
921 | ||
34dc7c2f | 922 | void |
572e2857 | 923 | dnode_special_close(dnode_handle_t *dnh) |
34dc7c2f | 924 | { |
572e2857 BB |
925 | dnode_t *dn = dnh->dnh_dnode; |
926 | ||
34dc7c2f BB |
927 | /* |
928 | * Wait for final references to the dnode to clear. This can | |
929 | * only happen if the arc is asyncronously evicting state that | |
930 | * has a hold on this dnode while we are trying to evict this | |
931 | * dnode. | |
932 | */ | |
933 | while (refcount_count(&dn->dn_holds) > 0) | |
934 | delay(1); | |
572e2857 BB |
935 | zrl_add(&dnh->dnh_zrlock); |
936 | dnode_destroy(dn); /* implicit zrl_remove() */ | |
937 | zrl_destroy(&dnh->dnh_zrlock); | |
938 | dnh->dnh_dnode = NULL; | |
34dc7c2f BB |
939 | } |
940 | ||
941 | dnode_t * | |
572e2857 BB |
942 | dnode_special_open(objset_t *os, dnode_phys_t *dnp, uint64_t object, |
943 | dnode_handle_t *dnh) | |
34dc7c2f | 944 | { |
572e2857 BB |
945 | dnode_t *dn = dnode_create(os, dnp, NULL, object, dnh); |
946 | dnh->dnh_dnode = dn; | |
947 | zrl_init(&dnh->dnh_zrlock); | |
34dc7c2f BB |
948 | DNODE_VERIFY(dn); |
949 | return (dn); | |
950 | } | |
951 | ||
952 | static void | |
953 | dnode_buf_pageout(dmu_buf_t *db, void *arg) | |
954 | { | |
572e2857 | 955 | dnode_children_t *children_dnodes = arg; |
34dc7c2f BB |
956 | int i; |
957 | int epb = db->db_size >> DNODE_SHIFT; | |
958 | ||
572e2857 BB |
959 | ASSERT(epb == children_dnodes->dnc_count); |
960 | ||
34dc7c2f | 961 | for (i = 0; i < epb; i++) { |
572e2857 BB |
962 | dnode_handle_t *dnh = &children_dnodes->dnc_children[i]; |
963 | dnode_t *dn; | |
34dc7c2f | 964 | |
572e2857 BB |
965 | /* |
966 | * The dnode handle lock guards against the dnode moving to | |
967 | * another valid address, so there is no need here to guard | |
968 | * against changes to or from NULL. | |
969 | */ | |
970 | if (dnh->dnh_dnode == NULL) { | |
971 | zrl_destroy(&dnh->dnh_zrlock); | |
34dc7c2f | 972 | continue; |
572e2857 BB |
973 | } |
974 | ||
975 | zrl_add(&dnh->dnh_zrlock); | |
976 | dn = dnh->dnh_dnode; | |
34dc7c2f BB |
977 | /* |
978 | * If there are holds on this dnode, then there should | |
979 | * be holds on the dnode's containing dbuf as well; thus | |
572e2857 | 980 | * it wouldn't be eligible for eviction and this function |
34dc7c2f BB |
981 | * would not have been called. |
982 | */ | |
983 | ASSERT(refcount_is_zero(&dn->dn_holds)); | |
34dc7c2f BB |
984 | ASSERT(refcount_is_zero(&dn->dn_tx_holds)); |
985 | ||
572e2857 BB |
986 | dnode_destroy(dn); /* implicit zrl_remove() */ |
987 | zrl_destroy(&dnh->dnh_zrlock); | |
988 | dnh->dnh_dnode = NULL; | |
34dc7c2f | 989 | } |
572e2857 BB |
990 | kmem_free(children_dnodes, sizeof (dnode_children_t) + |
991 | (epb - 1) * sizeof (dnode_handle_t)); | |
34dc7c2f BB |
992 | } |
993 | ||
994 | /* | |
995 | * errors: | |
996 | * EINVAL - invalid object number. | |
997 | * EIO - i/o error. | |
998 | * succeeds even for free dnodes. | |
999 | */ | |
1000 | int | |
428870ff | 1001 | dnode_hold_impl(objset_t *os, uint64_t object, int flag, |
34dc7c2f BB |
1002 | void *tag, dnode_t **dnp) |
1003 | { | |
1004 | int epb, idx, err; | |
1005 | int drop_struct_lock = FALSE; | |
1006 | int type; | |
1007 | uint64_t blk; | |
1008 | dnode_t *mdn, *dn; | |
1009 | dmu_buf_impl_t *db; | |
572e2857 BB |
1010 | dnode_children_t *children_dnodes; |
1011 | dnode_handle_t *dnh; | |
34dc7c2f | 1012 | |
b128c09f BB |
1013 | /* |
1014 | * If you are holding the spa config lock as writer, you shouldn't | |
428870ff BB |
1015 | * be asking the DMU to do *anything* unless it's the root pool |
1016 | * which may require us to read from the root filesystem while | |
1017 | * holding some (not all) of the locks as writer. | |
b128c09f | 1018 | */ |
428870ff BB |
1019 | ASSERT(spa_config_held(os->os_spa, SCL_ALL, RW_WRITER) == 0 || |
1020 | (spa_is_root(os->os_spa) && | |
572e2857 | 1021 | spa_config_held(os->os_spa, SCL_STATE, RW_WRITER))); |
b128c09f | 1022 | |
9babb374 BB |
1023 | if (object == DMU_USERUSED_OBJECT || object == DMU_GROUPUSED_OBJECT) { |
1024 | dn = (object == DMU_USERUSED_OBJECT) ? | |
572e2857 | 1025 | DMU_USERUSED_DNODE(os) : DMU_GROUPUSED_DNODE(os); |
9babb374 BB |
1026 | if (dn == NULL) |
1027 | return (ENOENT); | |
1028 | type = dn->dn_type; | |
1029 | if ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) | |
1030 | return (ENOENT); | |
1031 | if ((flag & DNODE_MUST_BE_FREE) && type != DMU_OT_NONE) | |
1032 | return (EEXIST); | |
1033 | DNODE_VERIFY(dn); | |
1034 | (void) refcount_add(&dn->dn_holds, tag); | |
1035 | *dnp = dn; | |
1036 | return (0); | |
1037 | } | |
1038 | ||
34dc7c2f BB |
1039 | if (object == 0 || object >= DN_MAX_OBJECT) |
1040 | return (EINVAL); | |
1041 | ||
572e2857 BB |
1042 | mdn = DMU_META_DNODE(os); |
1043 | ASSERT(mdn->dn_object == DMU_META_DNODE_OBJECT); | |
34dc7c2f BB |
1044 | |
1045 | DNODE_VERIFY(mdn); | |
1046 | ||
1047 | if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) { | |
1048 | rw_enter(&mdn->dn_struct_rwlock, RW_READER); | |
1049 | drop_struct_lock = TRUE; | |
1050 | } | |
1051 | ||
1052 | blk = dbuf_whichblock(mdn, object * sizeof (dnode_phys_t)); | |
1053 | ||
1054 | db = dbuf_hold(mdn, blk, FTAG); | |
1055 | if (drop_struct_lock) | |
1056 | rw_exit(&mdn->dn_struct_rwlock); | |
1057 | if (db == NULL) | |
1058 | return (EIO); | |
1059 | err = dbuf_read(db, NULL, DB_RF_CANFAIL); | |
1060 | if (err) { | |
1061 | dbuf_rele(db, FTAG); | |
1062 | return (err); | |
1063 | } | |
1064 | ||
1065 | ASSERT3U(db->db.db_size, >=, 1<<DNODE_SHIFT); | |
1066 | epb = db->db.db_size >> DNODE_SHIFT; | |
1067 | ||
1068 | idx = object & (epb-1); | |
1069 | ||
572e2857 | 1070 | ASSERT(DB_DNODE(db)->dn_type == DMU_OT_DNODE); |
34dc7c2f BB |
1071 | children_dnodes = dmu_buf_get_user(&db->db); |
1072 | if (children_dnodes == NULL) { | |
572e2857 BB |
1073 | int i; |
1074 | dnode_children_t *winner; | |
1075 | children_dnodes = kmem_alloc(sizeof (dnode_children_t) + | |
abd39a82 BB |
1076 | (epb - 1) * sizeof (dnode_handle_t), |
1077 | KM_PUSHPAGE | KM_NODEBUG); | |
572e2857 BB |
1078 | children_dnodes->dnc_count = epb; |
1079 | dnh = &children_dnodes->dnc_children[0]; | |
1080 | for (i = 0; i < epb; i++) { | |
1081 | zrl_init(&dnh[i].dnh_zrlock); | |
1082 | dnh[i].dnh_dnode = NULL; | |
1083 | } | |
c65aa5b2 BB |
1084 | if ((winner = dmu_buf_set_user(&db->db, children_dnodes, NULL, |
1085 | dnode_buf_pageout))) { | |
572e2857 BB |
1086 | kmem_free(children_dnodes, sizeof (dnode_children_t) + |
1087 | (epb - 1) * sizeof (dnode_handle_t)); | |
34dc7c2f BB |
1088 | children_dnodes = winner; |
1089 | } | |
1090 | } | |
572e2857 | 1091 | ASSERT(children_dnodes->dnc_count == epb); |
34dc7c2f | 1092 | |
572e2857 BB |
1093 | dnh = &children_dnodes->dnc_children[idx]; |
1094 | zrl_add(&dnh->dnh_zrlock); | |
1095 | if ((dn = dnh->dnh_dnode) == NULL) { | |
1096 | dnode_phys_t *phys = (dnode_phys_t *)db->db.db_data+idx; | |
34dc7c2f BB |
1097 | dnode_t *winner; |
1098 | ||
572e2857 BB |
1099 | dn = dnode_create(os, phys, db, object, dnh); |
1100 | winner = atomic_cas_ptr(&dnh->dnh_dnode, NULL, dn); | |
34dc7c2f | 1101 | if (winner != NULL) { |
572e2857 BB |
1102 | zrl_add(&dnh->dnh_zrlock); |
1103 | dnode_destroy(dn); /* implicit zrl_remove() */ | |
34dc7c2f BB |
1104 | dn = winner; |
1105 | } | |
1106 | } | |
1107 | ||
1108 | mutex_enter(&dn->dn_mtx); | |
1109 | type = dn->dn_type; | |
1110 | if (dn->dn_free_txg || | |
1111 | ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) || | |
9babb374 | 1112 | ((flag & DNODE_MUST_BE_FREE) && |
428870ff | 1113 | (type != DMU_OT_NONE || !refcount_is_zero(&dn->dn_holds)))) { |
34dc7c2f | 1114 | mutex_exit(&dn->dn_mtx); |
572e2857 | 1115 | zrl_remove(&dnh->dnh_zrlock); |
34dc7c2f BB |
1116 | dbuf_rele(db, FTAG); |
1117 | return (type == DMU_OT_NONE ? ENOENT : EEXIST); | |
1118 | } | |
1119 | mutex_exit(&dn->dn_mtx); | |
1120 | ||
1121 | if (refcount_add(&dn->dn_holds, tag) == 1) | |
572e2857 BB |
1122 | dbuf_add_ref(db, dnh); |
1123 | /* Now we can rely on the hold to prevent the dnode from moving. */ | |
1124 | zrl_remove(&dnh->dnh_zrlock); | |
34dc7c2f BB |
1125 | |
1126 | DNODE_VERIFY(dn); | |
1127 | ASSERT3P(dn->dn_dbuf, ==, db); | |
1128 | ASSERT3U(dn->dn_object, ==, object); | |
1129 | dbuf_rele(db, FTAG); | |
1130 | ||
1131 | *dnp = dn; | |
1132 | return (0); | |
1133 | } | |
1134 | ||
1135 | /* | |
1136 | * Return held dnode if the object is allocated, NULL if not. | |
1137 | */ | |
1138 | int | |
428870ff | 1139 | dnode_hold(objset_t *os, uint64_t object, void *tag, dnode_t **dnp) |
34dc7c2f BB |
1140 | { |
1141 | return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp)); | |
1142 | } | |
1143 | ||
1144 | /* | |
1145 | * Can only add a reference if there is already at least one | |
1146 | * reference on the dnode. Returns FALSE if unable to add a | |
1147 | * new reference. | |
1148 | */ | |
1149 | boolean_t | |
1150 | dnode_add_ref(dnode_t *dn, void *tag) | |
1151 | { | |
1152 | mutex_enter(&dn->dn_mtx); | |
1153 | if (refcount_is_zero(&dn->dn_holds)) { | |
1154 | mutex_exit(&dn->dn_mtx); | |
1155 | return (FALSE); | |
1156 | } | |
1157 | VERIFY(1 < refcount_add(&dn->dn_holds, tag)); | |
1158 | mutex_exit(&dn->dn_mtx); | |
1159 | return (TRUE); | |
1160 | } | |
1161 | ||
1162 | void | |
1163 | dnode_rele(dnode_t *dn, void *tag) | |
1164 | { | |
1165 | uint64_t refs; | |
572e2857 BB |
1166 | /* Get while the hold prevents the dnode from moving. */ |
1167 | dmu_buf_impl_t *db = dn->dn_dbuf; | |
1168 | dnode_handle_t *dnh = dn->dn_handle; | |
34dc7c2f BB |
1169 | |
1170 | mutex_enter(&dn->dn_mtx); | |
1171 | refs = refcount_remove(&dn->dn_holds, tag); | |
1172 | mutex_exit(&dn->dn_mtx); | |
572e2857 BB |
1173 | |
1174 | /* | |
1175 | * It's unsafe to release the last hold on a dnode by dnode_rele() or | |
1176 | * indirectly by dbuf_rele() while relying on the dnode handle to | |
1177 | * prevent the dnode from moving, since releasing the last hold could | |
1178 | * result in the dnode's parent dbuf evicting its dnode handles. For | |
1179 | * that reason anyone calling dnode_rele() or dbuf_rele() without some | |
1180 | * other direct or indirect hold on the dnode must first drop the dnode | |
1181 | * handle. | |
1182 | */ | |
1183 | ASSERT(refs > 0 || dnh->dnh_zrlock.zr_owner != curthread); | |
1184 | ||
34dc7c2f | 1185 | /* NOTE: the DNODE_DNODE does not have a dn_dbuf */ |
572e2857 BB |
1186 | if (refs == 0 && db != NULL) { |
1187 | /* | |
1188 | * Another thread could add a hold to the dnode handle in | |
1189 | * dnode_hold_impl() while holding the parent dbuf. Since the | |
1190 | * hold on the parent dbuf prevents the handle from being | |
1191 | * destroyed, the hold on the handle is OK. We can't yet assert | |
1192 | * that the handle has zero references, but that will be | |
1193 | * asserted anyway when the handle gets destroyed. | |
1194 | */ | |
1195 | dbuf_rele(db, dnh); | |
1196 | } | |
34dc7c2f BB |
1197 | } |
1198 | ||
1199 | void | |
1200 | dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) | |
1201 | { | |
428870ff | 1202 | objset_t *os = dn->dn_objset; |
34dc7c2f BB |
1203 | uint64_t txg = tx->tx_txg; |
1204 | ||
9babb374 BB |
1205 | if (DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { |
1206 | dsl_dataset_dirty(os->os_dsl_dataset, tx); | |
34dc7c2f | 1207 | return; |
9babb374 | 1208 | } |
34dc7c2f BB |
1209 | |
1210 | DNODE_VERIFY(dn); | |
1211 | ||
1212 | #ifdef ZFS_DEBUG | |
1213 | mutex_enter(&dn->dn_mtx); | |
1214 | ASSERT(dn->dn_phys->dn_type || dn->dn_allocated_txg); | |
572e2857 | 1215 | ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); |
34dc7c2f BB |
1216 | mutex_exit(&dn->dn_mtx); |
1217 | #endif | |
1218 | ||
428870ff BB |
1219 | /* |
1220 | * Determine old uid/gid when necessary | |
1221 | */ | |
1222 | dmu_objset_userquota_get_ids(dn, B_TRUE, tx); | |
1223 | ||
34dc7c2f BB |
1224 | mutex_enter(&os->os_lock); |
1225 | ||
1226 | /* | |
1227 | * If we are already marked dirty, we're done. | |
1228 | */ | |
1229 | if (list_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) { | |
1230 | mutex_exit(&os->os_lock); | |
1231 | return; | |
1232 | } | |
1233 | ||
1234 | ASSERT(!refcount_is_zero(&dn->dn_holds) || list_head(&dn->dn_dbufs)); | |
1235 | ASSERT(dn->dn_datablksz != 0); | |
1236 | ASSERT3U(dn->dn_next_bonuslen[txg&TXG_MASK], ==, 0); | |
1237 | ASSERT3U(dn->dn_next_blksz[txg&TXG_MASK], ==, 0); | |
428870ff | 1238 | ASSERT3U(dn->dn_next_bonustype[txg&TXG_MASK], ==, 0); |
34dc7c2f BB |
1239 | |
1240 | dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n", | |
1241 | dn->dn_object, txg); | |
1242 | ||
1243 | if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) { | |
1244 | list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn); | |
1245 | } else { | |
1246 | list_insert_tail(&os->os_dirty_dnodes[txg&TXG_MASK], dn); | |
1247 | } | |
1248 | ||
1249 | mutex_exit(&os->os_lock); | |
1250 | ||
1251 | /* | |
1252 | * The dnode maintains a hold on its containing dbuf as | |
1253 | * long as there are holds on it. Each instantiated child | |
572e2857 | 1254 | * dbuf maintains a hold on the dnode. When the last child |
34dc7c2f BB |
1255 | * drops its hold, the dnode will drop its hold on the |
1256 | * containing dbuf. We add a "dirty hold" here so that the | |
1257 | * dnode will hang around after we finish processing its | |
1258 | * children. | |
1259 | */ | |
1260 | VERIFY(dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg)); | |
1261 | ||
1262 | (void) dbuf_dirty(dn->dn_dbuf, tx); | |
1263 | ||
1264 | dsl_dataset_dirty(os->os_dsl_dataset, tx); | |
1265 | } | |
1266 | ||
1267 | void | |
1268 | dnode_free(dnode_t *dn, dmu_tx_t *tx) | |
1269 | { | |
1270 | int txgoff = tx->tx_txg & TXG_MASK; | |
1271 | ||
1272 | dprintf("dn=%p txg=%llu\n", dn, tx->tx_txg); | |
1273 | ||
1274 | /* we should be the only holder... hopefully */ | |
1275 | /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */ | |
1276 | ||
1277 | mutex_enter(&dn->dn_mtx); | |
1278 | if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg) { | |
1279 | mutex_exit(&dn->dn_mtx); | |
1280 | return; | |
1281 | } | |
1282 | dn->dn_free_txg = tx->tx_txg; | |
1283 | mutex_exit(&dn->dn_mtx); | |
1284 | ||
1285 | /* | |
1286 | * If the dnode is already dirty, it needs to be moved from | |
1287 | * the dirty list to the free list. | |
1288 | */ | |
1289 | mutex_enter(&dn->dn_objset->os_lock); | |
1290 | if (list_link_active(&dn->dn_dirty_link[txgoff])) { | |
1291 | list_remove(&dn->dn_objset->os_dirty_dnodes[txgoff], dn); | |
1292 | list_insert_tail(&dn->dn_objset->os_free_dnodes[txgoff], dn); | |
1293 | mutex_exit(&dn->dn_objset->os_lock); | |
1294 | } else { | |
1295 | mutex_exit(&dn->dn_objset->os_lock); | |
1296 | dnode_setdirty(dn, tx); | |
1297 | } | |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * Try to change the block size for the indicated dnode. This can only | |
1302 | * succeed if there are no blocks allocated or dirty beyond first block | |
1303 | */ | |
1304 | int | |
1305 | dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) | |
1306 | { | |
1307 | dmu_buf_impl_t *db, *db_next; | |
b128c09f | 1308 | int err; |
34dc7c2f BB |
1309 | |
1310 | if (size == 0) | |
1311 | size = SPA_MINBLOCKSIZE; | |
1312 | if (size > SPA_MAXBLOCKSIZE) | |
1313 | size = SPA_MAXBLOCKSIZE; | |
1314 | else | |
1315 | size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); | |
1316 | ||
1317 | if (ibs == dn->dn_indblkshift) | |
1318 | ibs = 0; | |
1319 | ||
1320 | if (size >> SPA_MINBLOCKSHIFT == dn->dn_datablkszsec && ibs == 0) | |
1321 | return (0); | |
1322 | ||
1323 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1324 | ||
1325 | /* Check for any allocated blocks beyond the first */ | |
1326 | if (dn->dn_phys->dn_maxblkid != 0) | |
1327 | goto fail; | |
1328 | ||
1329 | mutex_enter(&dn->dn_dbufs_mtx); | |
1330 | for (db = list_head(&dn->dn_dbufs); db; db = db_next) { | |
1331 | db_next = list_next(&dn->dn_dbufs, db); | |
1332 | ||
428870ff BB |
1333 | if (db->db_blkid != 0 && db->db_blkid != DMU_BONUS_BLKID && |
1334 | db->db_blkid != DMU_SPILL_BLKID) { | |
34dc7c2f BB |
1335 | mutex_exit(&dn->dn_dbufs_mtx); |
1336 | goto fail; | |
1337 | } | |
1338 | } | |
1339 | mutex_exit(&dn->dn_dbufs_mtx); | |
1340 | ||
1341 | if (ibs && dn->dn_nlevels != 1) | |
1342 | goto fail; | |
1343 | ||
b128c09f BB |
1344 | /* resize the old block */ |
1345 | err = dbuf_hold_impl(dn, 0, 0, TRUE, FTAG, &db); | |
1346 | if (err == 0) | |
34dc7c2f | 1347 | dbuf_new_size(db, size, tx); |
b128c09f BB |
1348 | else if (err != ENOENT) |
1349 | goto fail; | |
34dc7c2f BB |
1350 | |
1351 | dnode_setdblksz(dn, size); | |
1352 | dnode_setdirty(dn, tx); | |
1353 | dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size; | |
1354 | if (ibs) { | |
1355 | dn->dn_indblkshift = ibs; | |
1356 | dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; | |
1357 | } | |
b128c09f | 1358 | /* rele after we have fixed the blocksize in the dnode */ |
34dc7c2f BB |
1359 | if (db) |
1360 | dbuf_rele(db, FTAG); | |
1361 | ||
1362 | rw_exit(&dn->dn_struct_rwlock); | |
1363 | return (0); | |
1364 | ||
1365 | fail: | |
1366 | rw_exit(&dn->dn_struct_rwlock); | |
1367 | return (ENOTSUP); | |
1368 | } | |
1369 | ||
b128c09f | 1370 | /* read-holding callers must not rely on the lock being continuously held */ |
34dc7c2f | 1371 | void |
b128c09f | 1372 | dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t have_read) |
34dc7c2f BB |
1373 | { |
1374 | uint64_t txgoff = tx->tx_txg & TXG_MASK; | |
34dc7c2f BB |
1375 | int epbs, new_nlevels; |
1376 | uint64_t sz; | |
1377 | ||
428870ff | 1378 | ASSERT(blkid != DMU_BONUS_BLKID); |
34dc7c2f | 1379 | |
b128c09f BB |
1380 | ASSERT(have_read ? |
1381 | RW_READ_HELD(&dn->dn_struct_rwlock) : | |
1382 | RW_WRITE_HELD(&dn->dn_struct_rwlock)); | |
1383 | ||
1384 | /* | |
1385 | * if we have a read-lock, check to see if we need to do any work | |
1386 | * before upgrading to a write-lock. | |
1387 | */ | |
1388 | if (have_read) { | |
1389 | if (blkid <= dn->dn_maxblkid) | |
1390 | return; | |
1391 | ||
1392 | if (!rw_tryupgrade(&dn->dn_struct_rwlock)) { | |
1393 | rw_exit(&dn->dn_struct_rwlock); | |
1394 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1395 | } | |
34dc7c2f BB |
1396 | } |
1397 | ||
1398 | if (blkid <= dn->dn_maxblkid) | |
1399 | goto out; | |
1400 | ||
1401 | dn->dn_maxblkid = blkid; | |
1402 | ||
1403 | /* | |
1404 | * Compute the number of levels necessary to support the new maxblkid. | |
1405 | */ | |
1406 | new_nlevels = 1; | |
1407 | epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
1408 | for (sz = dn->dn_nblkptr; | |
1409 | sz <= blkid && sz >= dn->dn_nblkptr; sz <<= epbs) | |
1410 | new_nlevels++; | |
1411 | ||
1412 | if (new_nlevels > dn->dn_nlevels) { | |
1413 | int old_nlevels = dn->dn_nlevels; | |
1414 | dmu_buf_impl_t *db; | |
1415 | list_t *list; | |
1416 | dbuf_dirty_record_t *new, *dr, *dr_next; | |
1417 | ||
1418 | dn->dn_nlevels = new_nlevels; | |
1419 | ||
1420 | ASSERT3U(new_nlevels, >, dn->dn_next_nlevels[txgoff]); | |
1421 | dn->dn_next_nlevels[txgoff] = new_nlevels; | |
1422 | ||
1423 | /* dirty the left indirects */ | |
1424 | db = dbuf_hold_level(dn, old_nlevels, 0, FTAG); | |
428870ff | 1425 | ASSERT(db != NULL); |
34dc7c2f BB |
1426 | new = dbuf_dirty(db, tx); |
1427 | dbuf_rele(db, FTAG); | |
1428 | ||
1429 | /* transfer the dirty records to the new indirect */ | |
1430 | mutex_enter(&dn->dn_mtx); | |
1431 | mutex_enter(&new->dt.di.dr_mtx); | |
1432 | list = &dn->dn_dirty_records[txgoff]; | |
1433 | for (dr = list_head(list); dr; dr = dr_next) { | |
1434 | dr_next = list_next(&dn->dn_dirty_records[txgoff], dr); | |
1435 | if (dr->dr_dbuf->db_level != new_nlevels-1 && | |
428870ff BB |
1436 | dr->dr_dbuf->db_blkid != DMU_BONUS_BLKID && |
1437 | dr->dr_dbuf->db_blkid != DMU_SPILL_BLKID) { | |
34dc7c2f BB |
1438 | ASSERT(dr->dr_dbuf->db_level == old_nlevels-1); |
1439 | list_remove(&dn->dn_dirty_records[txgoff], dr); | |
1440 | list_insert_tail(&new->dt.di.dr_children, dr); | |
1441 | dr->dr_parent = new; | |
1442 | } | |
1443 | } | |
1444 | mutex_exit(&new->dt.di.dr_mtx); | |
1445 | mutex_exit(&dn->dn_mtx); | |
1446 | } | |
1447 | ||
1448 | out: | |
b128c09f BB |
1449 | if (have_read) |
1450 | rw_downgrade(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
1451 | } |
1452 | ||
1453 | void | |
1454 | dnode_clear_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) | |
1455 | { | |
1456 | avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; | |
1457 | avl_index_t where; | |
1458 | free_range_t *rp; | |
1459 | free_range_t rp_tofind; | |
1460 | uint64_t endblk = blkid + nblks; | |
1461 | ||
1462 | ASSERT(MUTEX_HELD(&dn->dn_mtx)); | |
1463 | ASSERT(nblks <= UINT64_MAX - blkid); /* no overflow */ | |
1464 | ||
1465 | dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", | |
1466 | blkid, nblks, tx->tx_txg); | |
1467 | rp_tofind.fr_blkid = blkid; | |
1468 | rp = avl_find(tree, &rp_tofind, &where); | |
1469 | if (rp == NULL) | |
1470 | rp = avl_nearest(tree, where, AVL_BEFORE); | |
1471 | if (rp == NULL) | |
1472 | rp = avl_nearest(tree, where, AVL_AFTER); | |
1473 | ||
1474 | while (rp && (rp->fr_blkid <= blkid + nblks)) { | |
1475 | uint64_t fr_endblk = rp->fr_blkid + rp->fr_nblks; | |
1476 | free_range_t *nrp = AVL_NEXT(tree, rp); | |
1477 | ||
1478 | if (blkid <= rp->fr_blkid && endblk >= fr_endblk) { | |
1479 | /* clear this entire range */ | |
1480 | avl_remove(tree, rp); | |
1481 | kmem_free(rp, sizeof (free_range_t)); | |
1482 | } else if (blkid <= rp->fr_blkid && | |
1483 | endblk > rp->fr_blkid && endblk < fr_endblk) { | |
1484 | /* clear the beginning of this range */ | |
1485 | rp->fr_blkid = endblk; | |
1486 | rp->fr_nblks = fr_endblk - endblk; | |
1487 | } else if (blkid > rp->fr_blkid && blkid < fr_endblk && | |
1488 | endblk >= fr_endblk) { | |
1489 | /* clear the end of this range */ | |
1490 | rp->fr_nblks = blkid - rp->fr_blkid; | |
1491 | } else if (blkid > rp->fr_blkid && endblk < fr_endblk) { | |
1492 | /* clear a chunk out of this range */ | |
1493 | free_range_t *new_rp = | |
b8d06fca | 1494 | kmem_alloc(sizeof (free_range_t), KM_PUSHPAGE); |
34dc7c2f BB |
1495 | |
1496 | new_rp->fr_blkid = endblk; | |
1497 | new_rp->fr_nblks = fr_endblk - endblk; | |
1498 | avl_insert_here(tree, new_rp, rp, AVL_AFTER); | |
1499 | rp->fr_nblks = blkid - rp->fr_blkid; | |
1500 | } | |
1501 | /* there may be no overlap */ | |
1502 | rp = nrp; | |
1503 | } | |
1504 | } | |
1505 | ||
1506 | void | |
1507 | dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx) | |
1508 | { | |
1509 | dmu_buf_impl_t *db; | |
1510 | uint64_t blkoff, blkid, nblks; | |
b128c09f | 1511 | int blksz, blkshift, head, tail; |
34dc7c2f | 1512 | int trunc = FALSE; |
b128c09f | 1513 | int epbs; |
34dc7c2f BB |
1514 | |
1515 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1516 | blksz = dn->dn_datablksz; | |
b128c09f BB |
1517 | blkshift = dn->dn_datablkshift; |
1518 | epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; | |
34dc7c2f | 1519 | |
34dc7c2f BB |
1520 | if (len == -1ULL) { |
1521 | len = UINT64_MAX - off; | |
1522 | trunc = TRUE; | |
1523 | } | |
1524 | ||
1525 | /* | |
1526 | * First, block align the region to free: | |
1527 | */ | |
1528 | if (ISP2(blksz)) { | |
1529 | head = P2NPHASE(off, blksz); | |
1530 | blkoff = P2PHASE(off, blksz); | |
b128c09f BB |
1531 | if ((off >> blkshift) > dn->dn_maxblkid) |
1532 | goto out; | |
34dc7c2f BB |
1533 | } else { |
1534 | ASSERT(dn->dn_maxblkid == 0); | |
1535 | if (off == 0 && len >= blksz) { | |
b128c09f BB |
1536 | /* Freeing the whole block; fast-track this request */ |
1537 | blkid = 0; | |
1538 | nblks = 1; | |
1539 | goto done; | |
1540 | } else if (off >= blksz) { | |
1541 | /* Freeing past end-of-data */ | |
1542 | goto out; | |
34dc7c2f BB |
1543 | } else { |
1544 | /* Freeing part of the block. */ | |
1545 | head = blksz - off; | |
1546 | ASSERT3U(head, >, 0); | |
1547 | } | |
1548 | blkoff = off; | |
1549 | } | |
1550 | /* zero out any partial block data at the start of the range */ | |
1551 | if (head) { | |
1552 | ASSERT3U(blkoff + head, ==, blksz); | |
1553 | if (len < head) | |
1554 | head = len; | |
1555 | if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off), TRUE, | |
1556 | FTAG, &db) == 0) { | |
1557 | caddr_t data; | |
1558 | ||
1559 | /* don't dirty if it isn't on disk and isn't dirty */ | |
1560 | if (db->db_last_dirty || | |
1561 | (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { | |
1562 | rw_exit(&dn->dn_struct_rwlock); | |
1563 | dbuf_will_dirty(db, tx); | |
1564 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1565 | data = db->db.db_data; | |
1566 | bzero(data + blkoff, head); | |
1567 | } | |
1568 | dbuf_rele(db, FTAG); | |
1569 | } | |
1570 | off += head; | |
1571 | len -= head; | |
1572 | } | |
1573 | ||
1574 | /* If the range was less than one block, we're done */ | |
b128c09f | 1575 | if (len == 0) |
34dc7c2f BB |
1576 | goto out; |
1577 | ||
b128c09f BB |
1578 | /* If the remaining range is past end of file, we're done */ |
1579 | if ((off >> blkshift) > dn->dn_maxblkid) | |
1580 | goto out; | |
34dc7c2f | 1581 | |
b128c09f BB |
1582 | ASSERT(ISP2(blksz)); |
1583 | if (trunc) | |
1584 | tail = 0; | |
1585 | else | |
1586 | tail = P2PHASE(len, blksz); | |
1587 | ||
1588 | ASSERT3U(P2PHASE(off, blksz), ==, 0); | |
1589 | /* zero out any partial block data at the end of the range */ | |
1590 | if (tail) { | |
1591 | if (len < tail) | |
1592 | tail = len; | |
1593 | if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off+len), | |
1594 | TRUE, FTAG, &db) == 0) { | |
1595 | /* don't dirty if not on disk and not dirty */ | |
1596 | if (db->db_last_dirty || | |
1597 | (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { | |
1598 | rw_exit(&dn->dn_struct_rwlock); | |
1599 | dbuf_will_dirty(db, tx); | |
1600 | rw_enter(&dn->dn_struct_rwlock, RW_WRITER); | |
1601 | bzero(db->db.db_data, tail); | |
34dc7c2f | 1602 | } |
b128c09f | 1603 | dbuf_rele(db, FTAG); |
34dc7c2f | 1604 | } |
b128c09f BB |
1605 | len -= tail; |
1606 | } | |
34dc7c2f | 1607 | |
b128c09f BB |
1608 | /* If the range did not include a full block, we are done */ |
1609 | if (len == 0) | |
1610 | goto out; | |
1611 | ||
1612 | ASSERT(IS_P2ALIGNED(off, blksz)); | |
1613 | ASSERT(trunc || IS_P2ALIGNED(len, blksz)); | |
1614 | blkid = off >> blkshift; | |
1615 | nblks = len >> blkshift; | |
1616 | if (trunc) | |
1617 | nblks += 1; | |
1618 | ||
1619 | /* | |
1620 | * Read in and mark all the level-1 indirects dirty, | |
1621 | * so that they will stay in memory until syncing phase. | |
1622 | * Always dirty the first and last indirect to make sure | |
1623 | * we dirty all the partial indirects. | |
1624 | */ | |
1625 | if (dn->dn_nlevels > 1) { | |
1626 | uint64_t i, first, last; | |
1627 | int shift = epbs + dn->dn_datablkshift; | |
1628 | ||
1629 | first = blkid >> epbs; | |
c65aa5b2 | 1630 | if ((db = dbuf_hold_level(dn, 1, first, FTAG))) { |
34dc7c2f BB |
1631 | dbuf_will_dirty(db, tx); |
1632 | dbuf_rele(db, FTAG); | |
1633 | } | |
b128c09f BB |
1634 | if (trunc) |
1635 | last = dn->dn_maxblkid >> epbs; | |
1636 | else | |
1637 | last = (blkid + nblks - 1) >> epbs; | |
1638 | if (last > first && (db = dbuf_hold_level(dn, 1, last, FTAG))) { | |
34dc7c2f BB |
1639 | dbuf_will_dirty(db, tx); |
1640 | dbuf_rele(db, FTAG); | |
1641 | } | |
b128c09f BB |
1642 | for (i = first + 1; i < last; i++) { |
1643 | uint64_t ibyte = i << shift; | |
1644 | int err; | |
1645 | ||
1646 | err = dnode_next_offset(dn, | |
1647 | DNODE_FIND_HAVELOCK, &ibyte, 1, 1, 0); | |
1648 | i = ibyte >> shift; | |
1649 | if (err == ESRCH || i >= last) | |
1650 | break; | |
1651 | ASSERT(err == 0); | |
1652 | db = dbuf_hold_level(dn, 1, i, FTAG); | |
1653 | if (db) { | |
1654 | dbuf_will_dirty(db, tx); | |
1655 | dbuf_rele(db, FTAG); | |
1656 | } | |
1657 | } | |
34dc7c2f | 1658 | } |
b128c09f BB |
1659 | done: |
1660 | /* | |
1661 | * Add this range to the dnode range list. | |
1662 | * We will finish up this free operation in the syncing phase. | |
1663 | */ | |
34dc7c2f BB |
1664 | mutex_enter(&dn->dn_mtx); |
1665 | dnode_clear_range(dn, blkid, nblks, tx); | |
1666 | { | |
1667 | free_range_t *rp, *found; | |
1668 | avl_index_t where; | |
1669 | avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK]; | |
1670 | ||
1671 | /* Add new range to dn_ranges */ | |
b8d06fca | 1672 | rp = kmem_alloc(sizeof (free_range_t), KM_PUSHPAGE); |
34dc7c2f BB |
1673 | rp->fr_blkid = blkid; |
1674 | rp->fr_nblks = nblks; | |
1675 | found = avl_find(tree, rp, &where); | |
1676 | ASSERT(found == NULL); | |
1677 | avl_insert(tree, rp, where); | |
1678 | dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", | |
1679 | blkid, nblks, tx->tx_txg); | |
1680 | } | |
1681 | mutex_exit(&dn->dn_mtx); | |
1682 | ||
b128c09f | 1683 | dbuf_free_range(dn, blkid, blkid + nblks - 1, tx); |
34dc7c2f BB |
1684 | dnode_setdirty(dn, tx); |
1685 | out: | |
b128c09f BB |
1686 | if (trunc && dn->dn_maxblkid >= (off >> blkshift)) |
1687 | dn->dn_maxblkid = (off >> blkshift ? (off >> blkshift) - 1 : 0); | |
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 | { | |
1710 | free_range_t range_tofind; | |
1711 | void *dp = spa_get_dsl(dn->dn_objset->os_spa); | |
1712 | int i; | |
1713 | ||
428870ff | 1714 | if (blkid == DMU_BONUS_BLKID) |
34dc7c2f BB |
1715 | return (FALSE); |
1716 | ||
1717 | /* | |
1718 | * If we're in the process of opening the pool, dp will not be | |
1719 | * set yet, but there shouldn't be anything dirty. | |
1720 | */ | |
1721 | if (dp == NULL) | |
1722 | return (FALSE); | |
1723 | ||
1724 | if (dn->dn_free_txg) | |
1725 | return (TRUE); | |
1726 | ||
428870ff BB |
1727 | if (blkid == DMU_SPILL_BLKID) |
1728 | return (dnode_spill_freed(dn)); | |
1729 | ||
34dc7c2f BB |
1730 | range_tofind.fr_blkid = blkid; |
1731 | mutex_enter(&dn->dn_mtx); | |
1732 | for (i = 0; i < TXG_SIZE; i++) { | |
1733 | free_range_t *range_found; | |
1734 | avl_index_t idx; | |
1735 | ||
1736 | range_found = avl_find(&dn->dn_ranges[i], &range_tofind, &idx); | |
1737 | if (range_found) { | |
1738 | ASSERT(range_found->fr_nblks > 0); | |
1739 | break; | |
1740 | } | |
1741 | range_found = avl_nearest(&dn->dn_ranges[i], idx, AVL_BEFORE); | |
1742 | if (range_found && | |
1743 | range_found->fr_blkid + range_found->fr_nblks > blkid) | |
1744 | break; | |
1745 | } | |
1746 | mutex_exit(&dn->dn_mtx); | |
1747 | return (i < TXG_SIZE); | |
1748 | } | |
1749 | ||
1750 | /* call from syncing context when we actually write/free space for this dnode */ | |
1751 | void | |
1752 | dnode_diduse_space(dnode_t *dn, int64_t delta) | |
1753 | { | |
1754 | uint64_t space; | |
1755 | dprintf_dnode(dn, "dn=%p dnp=%p used=%llu delta=%lld\n", | |
1756 | dn, dn->dn_phys, | |
1757 | (u_longlong_t)dn->dn_phys->dn_used, | |
1758 | (longlong_t)delta); | |
1759 | ||
1760 | mutex_enter(&dn->dn_mtx); | |
1761 | space = DN_USED_BYTES(dn->dn_phys); | |
1762 | if (delta > 0) { | |
1763 | ASSERT3U(space + delta, >=, space); /* no overflow */ | |
1764 | } else { | |
1765 | ASSERT3U(space, >=, -delta); /* no underflow */ | |
1766 | } | |
1767 | space += delta; | |
1768 | if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_DNODE_BYTES) { | |
1769 | ASSERT((dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) == 0); | |
1770 | ASSERT3U(P2PHASE(space, 1<<DEV_BSHIFT), ==, 0); | |
1771 | dn->dn_phys->dn_used = space >> DEV_BSHIFT; | |
1772 | } else { | |
1773 | dn->dn_phys->dn_used = space; | |
1774 | dn->dn_phys->dn_flags |= DNODE_FLAG_USED_BYTES; | |
1775 | } | |
1776 | mutex_exit(&dn->dn_mtx); | |
1777 | } | |
1778 | ||
1779 | /* | |
1780 | * Call when we think we're going to write/free space in open context. | |
1781 | * Be conservative (ie. OK to write less than this or free more than | |
1782 | * this, but don't write more or free less). | |
1783 | */ | |
1784 | void | |
1785 | dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx) | |
1786 | { | |
428870ff | 1787 | objset_t *os = dn->dn_objset; |
34dc7c2f BB |
1788 | dsl_dataset_t *ds = os->os_dsl_dataset; |
1789 | ||
1790 | if (space > 0) | |
1791 | space = spa_get_asize(os->os_spa, space); | |
1792 | ||
1793 | if (ds) | |
1794 | dsl_dir_willuse_space(ds->ds_dir, space, tx); | |
1795 | ||
1796 | dmu_tx_willuse_space(tx, space); | |
1797 | } | |
1798 | ||
45d1cae3 BB |
1799 | /* |
1800 | * This function scans a block at the indicated "level" looking for | |
1801 | * a hole or data (depending on 'flags'). If level > 0, then we are | |
1802 | * scanning an indirect block looking at its pointers. If level == 0, | |
1803 | * then we are looking at a block of dnodes. If we don't find what we | |
1804 | * are looking for in the block, we return ESRCH. Otherwise, return | |
1805 | * with *offset pointing to the beginning (if searching forwards) or | |
1806 | * end (if searching backwards) of the range covered by the block | |
1807 | * pointer we matched on (or dnode). | |
1808 | * | |
1809 | * The basic search algorithm used below by dnode_next_offset() is to | |
1810 | * use this function to search up the block tree (widen the search) until | |
1811 | * we find something (i.e., we don't return ESRCH) and then search back | |
1812 | * down the tree (narrow the search) until we reach our original search | |
1813 | * level. | |
1814 | */ | |
34dc7c2f | 1815 | static int |
b128c09f | 1816 | dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset, |
34dc7c2f BB |
1817 | int lvl, uint64_t blkfill, uint64_t txg) |
1818 | { | |
1819 | dmu_buf_impl_t *db = NULL; | |
1820 | void *data = NULL; | |
1821 | uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; | |
1822 | uint64_t epb = 1ULL << epbs; | |
1823 | uint64_t minfill, maxfill; | |
b128c09f BB |
1824 | boolean_t hole; |
1825 | int i, inc, error, span; | |
34dc7c2f BB |
1826 | |
1827 | dprintf("probing object %llu offset %llx level %d of %u\n", | |
1828 | dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels); | |
1829 | ||
9babb374 | 1830 | hole = ((flags & DNODE_FIND_HOLE) != 0); |
b128c09f BB |
1831 | inc = (flags & DNODE_FIND_BACKWARDS) ? -1 : 1; |
1832 | ASSERT(txg == 0 || !hole); | |
1833 | ||
34dc7c2f BB |
1834 | if (lvl == dn->dn_phys->dn_nlevels) { |
1835 | error = 0; | |
1836 | epb = dn->dn_phys->dn_nblkptr; | |
1837 | data = dn->dn_phys->dn_blkptr; | |
1838 | } else { | |
1839 | uint64_t blkid = dbuf_whichblock(dn, *offset) >> (epbs * lvl); | |
1840 | error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FTAG, &db); | |
1841 | if (error) { | |
b128c09f BB |
1842 | if (error != ENOENT) |
1843 | return (error); | |
1844 | if (hole) | |
1845 | return (0); | |
1846 | /* | |
1847 | * This can only happen when we are searching up | |
1848 | * the block tree for data. We don't really need to | |
1849 | * adjust the offset, as we will just end up looking | |
1850 | * at the pointer to this block in its parent, and its | |
1851 | * going to be unallocated, so we will skip over it. | |
1852 | */ | |
1853 | return (ESRCH); | |
34dc7c2f BB |
1854 | } |
1855 | error = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_HAVESTRUCT); | |
1856 | if (error) { | |
1857 | dbuf_rele(db, FTAG); | |
1858 | return (error); | |
1859 | } | |
1860 | data = db->db.db_data; | |
1861 | } | |
1862 | ||
1863 | if (db && txg && | |
1864 | (db->db_blkptr == NULL || db->db_blkptr->blk_birth <= txg)) { | |
b128c09f BB |
1865 | /* |
1866 | * This can only happen when we are searching up the tree | |
1867 | * and these conditions mean that we need to keep climbing. | |
1868 | */ | |
34dc7c2f BB |
1869 | error = ESRCH; |
1870 | } else if (lvl == 0) { | |
1871 | dnode_phys_t *dnp = data; | |
1872 | span = DNODE_SHIFT; | |
1873 | ASSERT(dn->dn_type == DMU_OT_DNODE); | |
1874 | ||
b128c09f BB |
1875 | for (i = (*offset >> span) & (blkfill - 1); |
1876 | i >= 0 && i < blkfill; i += inc) { | |
9babb374 | 1877 | if ((dnp[i].dn_type == DMU_OT_NONE) == hole) |
34dc7c2f | 1878 | break; |
b128c09f | 1879 | *offset += (1ULL << span) * inc; |
34dc7c2f | 1880 | } |
b128c09f | 1881 | if (i < 0 || i == blkfill) |
34dc7c2f BB |
1882 | error = ESRCH; |
1883 | } else { | |
1884 | blkptr_t *bp = data; | |
45d1cae3 | 1885 | uint64_t start = *offset; |
34dc7c2f BB |
1886 | span = (lvl - 1) * epbs + dn->dn_datablkshift; |
1887 | minfill = 0; | |
1888 | maxfill = blkfill << ((lvl - 1) * epbs); | |
1889 | ||
1890 | if (hole) | |
1891 | maxfill--; | |
1892 | else | |
1893 | minfill++; | |
1894 | ||
45d1cae3 BB |
1895 | *offset = *offset >> span; |
1896 | for (i = BF64_GET(*offset, 0, epbs); | |
b128c09f | 1897 | i >= 0 && i < epb; i += inc) { |
34dc7c2f BB |
1898 | if (bp[i].blk_fill >= minfill && |
1899 | bp[i].blk_fill <= maxfill && | |
b128c09f | 1900 | (hole || bp[i].blk_birth > txg)) |
34dc7c2f | 1901 | break; |
45d1cae3 BB |
1902 | if (inc > 0 || *offset > 0) |
1903 | *offset += inc; | |
1904 | } | |
1905 | *offset = *offset << span; | |
1906 | if (inc < 0) { | |
1907 | /* traversing backwards; position offset at the end */ | |
1908 | ASSERT3U(*offset, <=, start); | |
1909 | *offset = MIN(*offset + (1ULL << span) - 1, start); | |
1910 | } else if (*offset < start) { | |
1911 | *offset = start; | |
34dc7c2f | 1912 | } |
45d1cae3 | 1913 | if (i < 0 || i >= epb) |
34dc7c2f BB |
1914 | error = ESRCH; |
1915 | } | |
1916 | ||
1917 | if (db) | |
1918 | dbuf_rele(db, FTAG); | |
1919 | ||
1920 | return (error); | |
1921 | } | |
1922 | ||
1923 | /* | |
1924 | * Find the next hole, data, or sparse region at or after *offset. | |
1925 | * The value 'blkfill' tells us how many items we expect to find | |
1926 | * in an L0 data block; this value is 1 for normal objects, | |
1927 | * DNODES_PER_BLOCK for the meta dnode, and some fraction of | |
1928 | * DNODES_PER_BLOCK when searching for sparse regions thereof. | |
1929 | * | |
1930 | * Examples: | |
1931 | * | |
b128c09f BB |
1932 | * dnode_next_offset(dn, flags, offset, 1, 1, 0); |
1933 | * Finds the next/previous hole/data in a file. | |
34dc7c2f BB |
1934 | * Used in dmu_offset_next(). |
1935 | * | |
b128c09f | 1936 | * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg); |
34dc7c2f BB |
1937 | * Finds the next free/allocated dnode an objset's meta-dnode. |
1938 | * Only finds objects that have new contents since txg (ie. | |
1939 | * bonus buffer changes and content removal are ignored). | |
1940 | * Used in dmu_object_next(). | |
1941 | * | |
b128c09f | 1942 | * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0); |
34dc7c2f BB |
1943 | * Finds the next L2 meta-dnode bp that's at most 1/4 full. |
1944 | * Used in dmu_object_alloc(). | |
1945 | */ | |
1946 | int | |
b128c09f | 1947 | dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset, |
34dc7c2f BB |
1948 | int minlvl, uint64_t blkfill, uint64_t txg) |
1949 | { | |
b128c09f | 1950 | uint64_t initial_offset = *offset; |
34dc7c2f BB |
1951 | int lvl, maxlvl; |
1952 | int error = 0; | |
34dc7c2f | 1953 | |
b128c09f BB |
1954 | if (!(flags & DNODE_FIND_HAVELOCK)) |
1955 | rw_enter(&dn->dn_struct_rwlock, RW_READER); | |
34dc7c2f BB |
1956 | |
1957 | if (dn->dn_phys->dn_nlevels == 0) { | |
b128c09f BB |
1958 | error = ESRCH; |
1959 | goto out; | |
34dc7c2f BB |
1960 | } |
1961 | ||
1962 | if (dn->dn_datablkshift == 0) { | |
1963 | if (*offset < dn->dn_datablksz) { | |
b128c09f | 1964 | if (flags & DNODE_FIND_HOLE) |
34dc7c2f BB |
1965 | *offset = dn->dn_datablksz; |
1966 | } else { | |
1967 | error = ESRCH; | |
1968 | } | |
b128c09f | 1969 | goto out; |
34dc7c2f BB |
1970 | } |
1971 | ||
1972 | maxlvl = dn->dn_phys->dn_nlevels; | |
1973 | ||
1974 | for (lvl = minlvl; lvl <= maxlvl; lvl++) { | |
1975 | error = dnode_next_offset_level(dn, | |
b128c09f | 1976 | flags, offset, lvl, blkfill, txg); |
34dc7c2f BB |
1977 | if (error != ESRCH) |
1978 | break; | |
1979 | } | |
1980 | ||
b128c09f | 1981 | while (error == 0 && --lvl >= minlvl) { |
34dc7c2f | 1982 | error = dnode_next_offset_level(dn, |
b128c09f | 1983 | flags, offset, lvl, blkfill, txg); |
34dc7c2f BB |
1984 | } |
1985 | ||
b128c09f BB |
1986 | if (error == 0 && (flags & DNODE_FIND_BACKWARDS ? |
1987 | initial_offset < *offset : initial_offset > *offset)) | |
34dc7c2f | 1988 | error = ESRCH; |
b128c09f BB |
1989 | out: |
1990 | if (!(flags & DNODE_FIND_HAVELOCK)) | |
1991 | rw_exit(&dn->dn_struct_rwlock); | |
34dc7c2f BB |
1992 | |
1993 | return (error); | |
1994 | } |