]>
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 | |
1d3ba0bf | 9 | * or https://opensource.org/licenses/CDDL-1.0. |
34dc7c2f BB |
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. |
d9b4bf06 | 23 | * Copyright (c) 2012, 2018 by Delphix. All rights reserved. |
0c66c32d | 24 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. |
34dc7c2f BB |
25 | */ |
26 | ||
34dc7c2f BB |
27 | /* |
28 | * This file contains the top half of the zfs directory structure | |
29 | * implementation. The bottom half is in zap_leaf.c. | |
30 | * | |
31 | * The zdir is an extendable hash data structure. There is a table of | |
32 | * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are | |
33 | * each a constant size and hold a variable number of directory entries. | |
34 | * The buckets (aka "leaf nodes") are implemented in zap_leaf.c. | |
35 | * | |
36 | * The pointer table holds a power of 2 number of pointers. | |
37 | * (1<<zap_t->zd_data->zd_phys->zd_prefix_len). The bucket pointed to | |
38 | * by the pointer at index i in the table holds entries whose hash value | |
39 | * has a zd_prefix_len - bit prefix | |
40 | */ | |
41 | ||
42 | #include <sys/spa.h> | |
43 | #include <sys/dmu.h> | |
44 | #include <sys/zfs_context.h> | |
45 | #include <sys/zfs_znode.h> | |
9babb374 | 46 | #include <sys/fs/zfs.h> |
34dc7c2f | 47 | #include <sys/zap.h> |
34dc7c2f BB |
48 | #include <sys/zap_impl.h> |
49 | #include <sys/zap_leaf.h> | |
50 | ||
d9b4bf06 MA |
51 | /* |
52 | * If zap_iterate_prefetch is set, we will prefetch the entire ZAP object | |
53 | * (all leaf blocks) when we start iterating over it. | |
54 | * | |
55 | * For zap_cursor_init(), the callers all intend to iterate through all the | |
56 | * entries. There are a few cases where an error (typically i/o error) could | |
57 | * cause it to bail out early. | |
58 | * | |
59 | * For zap_cursor_init_serialized(), there are callers that do the iteration | |
60 | * outside of ZFS. Typically they would iterate over everything, but we | |
61 | * don't have control of that. E.g. zfs_ioc_snapshot_list_next(), | |
62 | * zcp_snapshots_iter(), and other iterators over things in the MOS - these | |
63 | * are called by /sbin/zfs and channel programs. The other example is | |
64 | * zfs_readdir() which iterates over directory entries for the getdents() | |
65 | * syscall. /sbin/ls iterates to the end (unless it receives a signal), but | |
66 | * userland doesn't have to. | |
67 | * | |
68 | * Given that the ZAP entries aren't returned in a specific order, the only | |
69 | * legitimate use cases for partial iteration would be: | |
70 | * | |
71 | * 1. Pagination: e.g. you only want to display 100 entries at a time, so you | |
72 | * get the first 100 and then wait for the user to hit "next page", which | |
73 | * they may never do). | |
74 | * | |
75 | * 2. You want to know if there are more than X entries, without relying on | |
76 | * the zfs-specific implementation of the directory's st_size (which is | |
77 | * the number of entries). | |
78 | */ | |
18168da7 | 79 | static int zap_iterate_prefetch = B_TRUE; |
d9b4bf06 | 80 | |
34dc7c2f BB |
81 | int fzap_default_block_shift = 14; /* 16k blocksize */ |
82 | ||
34dc7c2f BB |
83 | static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks); |
84 | ||
34dc7c2f BB |
85 | void |
86 | fzap_byteswap(void *vbuf, size_t size) | |
87 | { | |
d2a12f9e | 88 | uint64_t block_type = *(uint64_t *)vbuf; |
34dc7c2f BB |
89 | |
90 | if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF)) | |
91 | zap_leaf_byteswap(vbuf, size); | |
92 | else { | |
93 | /* it's a ptrtbl block */ | |
94 | byteswap_uint64_array(vbuf, size); | |
95 | } | |
96 | } | |
97 | ||
98 | void | |
428870ff | 99 | fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags) |
34dc7c2f | 100 | { |
34dc7c2f BB |
101 | ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); |
102 | zap->zap_ismicro = FALSE; | |
103 | ||
39efbde7 GM |
104 | zap->zap_dbu.dbu_evict_func_sync = zap_evict_sync; |
105 | zap->zap_dbu.dbu_evict_func_async = NULL; | |
34dc7c2f | 106 | |
c17486b2 | 107 | mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, MUTEX_DEFAULT, 0); |
9bd274dd | 108 | zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1; |
34dc7c2f | 109 | |
d2a12f9e | 110 | zap_phys_t *zp = zap_f_phys(zap); |
34dc7c2f BB |
111 | /* |
112 | * explicitly zero it since it might be coming from an | |
113 | * initialized microzap | |
114 | */ | |
861166b0 | 115 | memset(zap->zap_dbuf->db_data, 0, zap->zap_dbuf->db_size); |
34dc7c2f BB |
116 | zp->zap_block_type = ZBT_HEADER; |
117 | zp->zap_magic = ZAP_MAGIC; | |
118 | ||
119 | zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap); | |
120 | ||
121 | zp->zap_freeblk = 2; /* block 1 will be the first leaf */ | |
122 | zp->zap_num_leafs = 1; | |
123 | zp->zap_num_entries = 0; | |
124 | zp->zap_salt = zap->zap_salt; | |
125 | zp->zap_normflags = zap->zap_normflags; | |
428870ff | 126 | zp->zap_flags = flags; |
34dc7c2f BB |
127 | |
128 | /* block 1 will be the first leaf */ | |
d2a12f9e | 129 | for (int i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++) |
34dc7c2f BB |
130 | ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1; |
131 | ||
132 | /* | |
133 | * set up block 1 - the first leaf | |
134 | */ | |
d2a12f9e MA |
135 | dmu_buf_t *db; |
136 | VERIFY0(dmu_buf_hold(zap->zap_objset, zap->zap_object, | |
428870ff | 137 | 1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH)); |
34dc7c2f BB |
138 | dmu_buf_will_dirty(db, tx); |
139 | ||
d2a12f9e | 140 | zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP); |
34dc7c2f | 141 | l->l_dbuf = db; |
34dc7c2f BB |
142 | |
143 | zap_leaf_init(l, zp->zap_normflags != 0); | |
144 | ||
145 | kmem_free(l, sizeof (zap_leaf_t)); | |
146 | dmu_buf_rele(db, FTAG); | |
147 | } | |
148 | ||
149 | static int | |
150 | zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx) | |
151 | { | |
152 | if (RW_WRITE_HELD(&zap->zap_rwlock)) | |
153 | return (1); | |
154 | if (rw_tryupgrade(&zap->zap_rwlock)) { | |
155 | dmu_buf_will_dirty(zap->zap_dbuf, tx); | |
156 | return (1); | |
157 | } | |
158 | return (0); | |
159 | } | |
160 | ||
161 | /* | |
162 | * Generic routines for dealing with the pointer & cookie tables. | |
163 | */ | |
164 | ||
165 | static int | |
166 | zap_table_grow(zap_t *zap, zap_table_phys_t *tbl, | |
167 | void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n), | |
168 | dmu_tx_t *tx) | |
169 | { | |
d2a12f9e | 170 | uint64_t newblk; |
34dc7c2f BB |
171 | int bs = FZAP_BLOCK_SHIFT(zap); |
172 | int hepb = 1<<(bs-4); | |
173 | /* hepb = half the number of entries in a block */ | |
174 | ||
175 | ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); | |
176 | ASSERT(tbl->zt_blk != 0); | |
177 | ASSERT(tbl->zt_numblks > 0); | |
178 | ||
179 | if (tbl->zt_nextblk != 0) { | |
180 | newblk = tbl->zt_nextblk; | |
181 | } else { | |
182 | newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2); | |
183 | tbl->zt_nextblk = newblk; | |
c99c9001 | 184 | ASSERT0(tbl->zt_blks_copied); |
fcff0f35 PD |
185 | dmu_prefetch(zap->zap_objset, zap->zap_object, 0, |
186 | tbl->zt_blk << bs, tbl->zt_numblks << bs, | |
187 | ZIO_PRIORITY_SYNC_READ); | |
34dc7c2f BB |
188 | } |
189 | ||
190 | /* | |
191 | * Copy the ptrtbl from the old to new location. | |
192 | */ | |
193 | ||
d2a12f9e MA |
194 | uint64_t b = tbl->zt_blks_copied; |
195 | dmu_buf_t *db_old; | |
196 | int err = dmu_buf_hold(zap->zap_objset, zap->zap_object, | |
428870ff | 197 | (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH); |
d2a12f9e | 198 | if (err != 0) |
34dc7c2f BB |
199 | return (err); |
200 | ||
201 | /* first half of entries in old[b] go to new[2*b+0] */ | |
d2a12f9e MA |
202 | dmu_buf_t *db_new; |
203 | VERIFY0(dmu_buf_hold(zap->zap_objset, zap->zap_object, | |
428870ff | 204 | (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH)); |
34dc7c2f BB |
205 | dmu_buf_will_dirty(db_new, tx); |
206 | transfer_func(db_old->db_data, db_new->db_data, hepb); | |
207 | dmu_buf_rele(db_new, FTAG); | |
208 | ||
209 | /* second half of entries in old[b] go to new[2*b+1] */ | |
d2a12f9e | 210 | VERIFY0(dmu_buf_hold(zap->zap_objset, zap->zap_object, |
428870ff | 211 | (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH)); |
34dc7c2f BB |
212 | dmu_buf_will_dirty(db_new, tx); |
213 | transfer_func((uint64_t *)db_old->db_data + hepb, | |
214 | db_new->db_data, hepb); | |
215 | dmu_buf_rele(db_new, FTAG); | |
216 | ||
217 | dmu_buf_rele(db_old, FTAG); | |
218 | ||
219 | tbl->zt_blks_copied++; | |
220 | ||
221 | dprintf("copied block %llu of %llu\n", | |
8e739b2c RE |
222 | (u_longlong_t)tbl->zt_blks_copied, |
223 | (u_longlong_t)tbl->zt_numblks); | |
34dc7c2f BB |
224 | |
225 | if (tbl->zt_blks_copied == tbl->zt_numblks) { | |
226 | (void) dmu_free_range(zap->zap_objset, zap->zap_object, | |
227 | tbl->zt_blk << bs, tbl->zt_numblks << bs, tx); | |
228 | ||
229 | tbl->zt_blk = newblk; | |
230 | tbl->zt_numblks *= 2; | |
231 | tbl->zt_shift++; | |
232 | tbl->zt_nextblk = 0; | |
233 | tbl->zt_blks_copied = 0; | |
234 | ||
29e57d15 | 235 | dprintf("finished; numblocks now %llu (%uk entries)\n", |
8e739b2c | 236 | (u_longlong_t)tbl->zt_numblks, 1<<(tbl->zt_shift-10)); |
34dc7c2f BB |
237 | } |
238 | ||
239 | return (0); | |
240 | } | |
241 | ||
242 | static int | |
243 | zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val, | |
244 | dmu_tx_t *tx) | |
245 | { | |
34dc7c2f | 246 | int bs = FZAP_BLOCK_SHIFT(zap); |
34dc7c2f BB |
247 | |
248 | ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); | |
249 | ASSERT(tbl->zt_blk != 0); | |
250 | ||
8e739b2c RE |
251 | dprintf("storing %llx at index %llx\n", (u_longlong_t)val, |
252 | (u_longlong_t)idx); | |
34dc7c2f | 253 | |
d2a12f9e MA |
254 | uint64_t blk = idx >> (bs-3); |
255 | uint64_t off = idx & ((1<<(bs-3))-1); | |
34dc7c2f | 256 | |
d2a12f9e MA |
257 | dmu_buf_t *db; |
258 | int err = dmu_buf_hold(zap->zap_objset, zap->zap_object, | |
428870ff | 259 | (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH); |
d2a12f9e | 260 | if (err != 0) |
34dc7c2f BB |
261 | return (err); |
262 | dmu_buf_will_dirty(db, tx); | |
263 | ||
264 | if (tbl->zt_nextblk != 0) { | |
265 | uint64_t idx2 = idx * 2; | |
266 | uint64_t blk2 = idx2 >> (bs-3); | |
267 | uint64_t off2 = idx2 & ((1<<(bs-3))-1); | |
268 | dmu_buf_t *db2; | |
269 | ||
270 | err = dmu_buf_hold(zap->zap_objset, zap->zap_object, | |
428870ff BB |
271 | (tbl->zt_nextblk + blk2) << bs, FTAG, &db2, |
272 | DMU_READ_NO_PREFETCH); | |
d2a12f9e | 273 | if (err != 0) { |
34dc7c2f BB |
274 | dmu_buf_rele(db, FTAG); |
275 | return (err); | |
276 | } | |
277 | dmu_buf_will_dirty(db2, tx); | |
278 | ((uint64_t *)db2->db_data)[off2] = val; | |
279 | ((uint64_t *)db2->db_data)[off2+1] = val; | |
280 | dmu_buf_rele(db2, FTAG); | |
281 | } | |
282 | ||
283 | ((uint64_t *)db->db_data)[off] = val; | |
284 | dmu_buf_rele(db, FTAG); | |
285 | ||
286 | return (0); | |
287 | } | |
288 | ||
289 | static int | |
290 | zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp) | |
291 | { | |
34dc7c2f BB |
292 | int bs = FZAP_BLOCK_SHIFT(zap); |
293 | ||
294 | ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); | |
295 | ||
d2a12f9e MA |
296 | uint64_t blk = idx >> (bs-3); |
297 | uint64_t off = idx & ((1<<(bs-3))-1); | |
34dc7c2f | 298 | |
2bce8049 MA |
299 | /* |
300 | * Note: this is equivalent to dmu_buf_hold(), but we use | |
301 | * _dnode_enter / _by_dnode because it's faster because we don't | |
302 | * have to hold the dnode. | |
303 | */ | |
d2a12f9e MA |
304 | dnode_t *dn = dmu_buf_dnode_enter(zap->zap_dbuf); |
305 | dmu_buf_t *db; | |
306 | int err = dmu_buf_hold_by_dnode(dn, | |
428870ff | 307 | (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH); |
2bce8049 | 308 | dmu_buf_dnode_exit(zap->zap_dbuf); |
d2a12f9e | 309 | if (err != 0) |
34dc7c2f BB |
310 | return (err); |
311 | *valp = ((uint64_t *)db->db_data)[off]; | |
312 | dmu_buf_rele(db, FTAG); | |
313 | ||
314 | if (tbl->zt_nextblk != 0) { | |
315 | /* | |
316 | * read the nextblk for the sake of i/o error checking, | |
317 | * so that zap_table_load() will catch errors for | |
318 | * zap_table_store. | |
319 | */ | |
320 | blk = (idx*2) >> (bs-3); | |
321 | ||
2bce8049 MA |
322 | dn = dmu_buf_dnode_enter(zap->zap_dbuf); |
323 | err = dmu_buf_hold_by_dnode(dn, | |
428870ff BB |
324 | (tbl->zt_nextblk + blk) << bs, FTAG, &db, |
325 | DMU_READ_NO_PREFETCH); | |
2bce8049 | 326 | dmu_buf_dnode_exit(zap->zap_dbuf); |
3a84951d WA |
327 | if (err == 0) |
328 | dmu_buf_rele(db, FTAG); | |
34dc7c2f BB |
329 | } |
330 | return (err); | |
331 | } | |
332 | ||
333 | /* | |
334 | * Routines for growing the ptrtbl. | |
335 | */ | |
336 | ||
337 | static void | |
338 | zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n) | |
339 | { | |
d2a12f9e | 340 | for (int i = 0; i < n; i++) { |
34dc7c2f | 341 | uint64_t lb = src[i]; |
d2a12f9e MA |
342 | dst[2 * i + 0] = lb; |
343 | dst[2 * i + 1] = lb; | |
34dc7c2f BB |
344 | } |
345 | } | |
346 | ||
347 | static int | |
348 | zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx) | |
349 | { | |
428870ff BB |
350 | /* |
351 | * The pointer table should never use more hash bits than we | |
352 | * have (otherwise we'd be using useless zero bits to index it). | |
353 | * If we are within 2 bits of running out, stop growing, since | |
354 | * this is already an aberrant condition. | |
355 | */ | |
d683ddbb | 356 | if (zap_f_phys(zap)->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2) |
2e528b49 | 357 | return (SET_ERROR(ENOSPC)); |
34dc7c2f | 358 | |
d683ddbb | 359 | if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) { |
34dc7c2f BB |
360 | /* |
361 | * We are outgrowing the "embedded" ptrtbl (the one | |
362 | * stored in the header block). Give it its own entire | |
363 | * block, which will double the size of the ptrtbl. | |
364 | */ | |
d683ddbb | 365 | ASSERT3U(zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==, |
34dc7c2f | 366 | ZAP_EMBEDDED_PTRTBL_SHIFT(zap)); |
d683ddbb | 367 | ASSERT0(zap_f_phys(zap)->zap_ptrtbl.zt_blk); |
34dc7c2f | 368 | |
d2a12f9e MA |
369 | uint64_t newblk = zap_allocate_blocks(zap, 1); |
370 | dmu_buf_t *db_new; | |
371 | int err = dmu_buf_hold(zap->zap_objset, zap->zap_object, | |
428870ff BB |
372 | newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new, |
373 | DMU_READ_NO_PREFETCH); | |
d2a12f9e | 374 | if (err != 0) |
34dc7c2f BB |
375 | return (err); |
376 | dmu_buf_will_dirty(db_new, tx); | |
377 | zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), | |
378 | db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap)); | |
379 | dmu_buf_rele(db_new, FTAG); | |
380 | ||
d683ddbb JG |
381 | zap_f_phys(zap)->zap_ptrtbl.zt_blk = newblk; |
382 | zap_f_phys(zap)->zap_ptrtbl.zt_numblks = 1; | |
383 | zap_f_phys(zap)->zap_ptrtbl.zt_shift++; | |
34dc7c2f | 384 | |
d683ddbb JG |
385 | ASSERT3U(1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==, |
386 | zap_f_phys(zap)->zap_ptrtbl.zt_numblks << | |
34dc7c2f BB |
387 | (FZAP_BLOCK_SHIFT(zap)-3)); |
388 | ||
389 | return (0); | |
390 | } else { | |
d683ddbb | 391 | return (zap_table_grow(zap, &zap_f_phys(zap)->zap_ptrtbl, |
34dc7c2f BB |
392 | zap_ptrtbl_transfer, tx)); |
393 | } | |
394 | } | |
395 | ||
396 | static void | |
397 | zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx) | |
398 | { | |
399 | dmu_buf_will_dirty(zap->zap_dbuf, tx); | |
400 | mutex_enter(&zap->zap_f.zap_num_entries_mtx); | |
d683ddbb JG |
401 | ASSERT(delta > 0 || zap_f_phys(zap)->zap_num_entries >= -delta); |
402 | zap_f_phys(zap)->zap_num_entries += delta; | |
34dc7c2f BB |
403 | mutex_exit(&zap->zap_f.zap_num_entries_mtx); |
404 | } | |
405 | ||
406 | static uint64_t | |
407 | zap_allocate_blocks(zap_t *zap, int nblocks) | |
408 | { | |
34dc7c2f | 409 | ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); |
d2a12f9e | 410 | uint64_t newblk = zap_f_phys(zap)->zap_freeblk; |
d683ddbb | 411 | zap_f_phys(zap)->zap_freeblk += nblocks; |
34dc7c2f BB |
412 | return (newblk); |
413 | } | |
414 | ||
0c66c32d | 415 | static void |
39efbde7 | 416 | zap_leaf_evict_sync(void *dbu) |
0c66c32d JG |
417 | { |
418 | zap_leaf_t *l = dbu; | |
419 | ||
420 | rw_destroy(&l->l_rwlock); | |
421 | kmem_free(l, sizeof (zap_leaf_t)); | |
422 | } | |
423 | ||
34dc7c2f BB |
424 | static zap_leaf_t * |
425 | zap_create_leaf(zap_t *zap, dmu_tx_t *tx) | |
426 | { | |
0c66c32d | 427 | zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP); |
34dc7c2f BB |
428 | |
429 | ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); | |
430 | ||
448d7aaa | 431 | rw_init(&l->l_rwlock, NULL, RW_NOLOCKDEP, NULL); |
34dc7c2f BB |
432 | rw_enter(&l->l_rwlock, RW_WRITER); |
433 | l->l_blkid = zap_allocate_blocks(zap, 1); | |
434 | l->l_dbuf = NULL; | |
34dc7c2f | 435 | |
d2a12f9e | 436 | VERIFY0(dmu_buf_hold(zap->zap_objset, zap->zap_object, |
428870ff BB |
437 | l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf, |
438 | DMU_READ_NO_PREFETCH)); | |
39efbde7 | 439 | dmu_buf_init_user(&l->l_dbu, zap_leaf_evict_sync, NULL, &l->l_dbuf); |
d2a12f9e | 440 | VERIFY3P(NULL, ==, dmu_buf_set_user(l->l_dbuf, &l->l_dbu)); |
34dc7c2f BB |
441 | dmu_buf_will_dirty(l->l_dbuf, tx); |
442 | ||
443 | zap_leaf_init(l, zap->zap_normflags != 0); | |
444 | ||
d683ddbb | 445 | zap_f_phys(zap)->zap_num_leafs++; |
34dc7c2f BB |
446 | |
447 | return (l); | |
448 | } | |
449 | ||
450 | int | |
451 | fzap_count(zap_t *zap, uint64_t *count) | |
452 | { | |
453 | ASSERT(!zap->zap_ismicro); | |
454 | mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */ | |
d683ddbb | 455 | *count = zap_f_phys(zap)->zap_num_entries; |
34dc7c2f BB |
456 | mutex_exit(&zap->zap_f.zap_num_entries_mtx); |
457 | return (0); | |
458 | } | |
459 | ||
460 | /* | |
461 | * Routines for obtaining zap_leaf_t's | |
462 | */ | |
463 | ||
464 | void | |
465 | zap_put_leaf(zap_leaf_t *l) | |
466 | { | |
467 | rw_exit(&l->l_rwlock); | |
468 | dmu_buf_rele(l->l_dbuf, NULL); | |
469 | } | |
470 | ||
34dc7c2f BB |
471 | static zap_leaf_t * |
472 | zap_open_leaf(uint64_t blkid, dmu_buf_t *db) | |
473 | { | |
34dc7c2f BB |
474 | ASSERT(blkid != 0); |
475 | ||
d2a12f9e | 476 | zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP); |
ef5319df | 477 | rw_init(&l->l_rwlock, NULL, RW_DEFAULT, NULL); |
34dc7c2f BB |
478 | rw_enter(&l->l_rwlock, RW_WRITER); |
479 | l->l_blkid = blkid; | |
9bd274dd | 480 | l->l_bs = highbit64(db->db_size) - 1; |
34dc7c2f | 481 | l->l_dbuf = db; |
34dc7c2f | 482 | |
39efbde7 | 483 | dmu_buf_init_user(&l->l_dbu, zap_leaf_evict_sync, NULL, &l->l_dbuf); |
d2a12f9e | 484 | zap_leaf_t *winner = dmu_buf_set_user(db, &l->l_dbu); |
34dc7c2f BB |
485 | |
486 | rw_exit(&l->l_rwlock); | |
487 | if (winner != NULL) { | |
488 | /* someone else set it first */ | |
39efbde7 | 489 | zap_leaf_evict_sync(&l->l_dbu); |
34dc7c2f BB |
490 | l = winner; |
491 | } | |
492 | ||
493 | /* | |
494 | * lhr_pad was previously used for the next leaf in the leaf | |
495 | * chain. There should be no chained leafs (as we have removed | |
496 | * support for them). | |
497 | */ | |
d683ddbb | 498 | ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1); |
34dc7c2f BB |
499 | |
500 | /* | |
501 | * There should be more hash entries than there can be | |
502 | * chunks to put in the hash table | |
503 | */ | |
504 | ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3); | |
505 | ||
506 | /* The chunks should begin at the end of the hash table */ | |
b8864a23 | 507 | ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==, (zap_leaf_chunk_t *) |
d683ddbb | 508 | &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]); |
34dc7c2f BB |
509 | |
510 | /* The chunks should end at the end of the block */ | |
511 | ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) - | |
d683ddbb | 512 | (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size); |
34dc7c2f BB |
513 | |
514 | return (l); | |
515 | } | |
516 | ||
517 | static int | |
518 | zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt, | |
519 | zap_leaf_t **lp) | |
520 | { | |
521 | dmu_buf_t *db; | |
34dc7c2f BB |
522 | |
523 | ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); | |
524 | ||
29572ccd CC |
525 | /* |
526 | * If system crashed just after dmu_free_long_range in zfs_rmnode, we | |
527 | * would be left with an empty xattr dir in delete queue. blkid=0 | |
528 | * would be passed in when doing zfs_purgedir. If that's the case we | |
529 | * should just return immediately. The underlying objects should | |
530 | * already be freed, so this should be perfectly fine. | |
531 | */ | |
532 | if (blkid == 0) | |
ecb2b7dc | 533 | return (SET_ERROR(ENOENT)); |
29572ccd | 534 | |
d2a12f9e | 535 | int bs = FZAP_BLOCK_SHIFT(zap); |
1c27024e | 536 | dnode_t *dn = dmu_buf_dnode_enter(zap->zap_dbuf); |
d2a12f9e | 537 | int err = dmu_buf_hold_by_dnode(dn, |
428870ff | 538 | blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH); |
2bce8049 | 539 | dmu_buf_dnode_exit(zap->zap_dbuf); |
d2a12f9e | 540 | if (err != 0) |
34dc7c2f BB |
541 | return (err); |
542 | ||
543 | ASSERT3U(db->db_object, ==, zap->zap_object); | |
544 | ASSERT3U(db->db_offset, ==, blkid << bs); | |
545 | ASSERT3U(db->db_size, ==, 1 << bs); | |
546 | ASSERT(blkid != 0); | |
547 | ||
d2a12f9e | 548 | zap_leaf_t *l = dmu_buf_get_user(db); |
34dc7c2f BB |
549 | |
550 | if (l == NULL) | |
551 | l = zap_open_leaf(blkid, db); | |
552 | ||
553 | rw_enter(&l->l_rwlock, lt); | |
554 | /* | |
d683ddbb | 555 | * Must lock before dirtying, otherwise zap_leaf_phys(l) could change, |
34dc7c2f BB |
556 | * causing ASSERT below to fail. |
557 | */ | |
558 | if (lt == RW_WRITER) | |
559 | dmu_buf_will_dirty(db, tx); | |
560 | ASSERT3U(l->l_blkid, ==, blkid); | |
561 | ASSERT3P(l->l_dbuf, ==, db); | |
d683ddbb JG |
562 | ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF); |
563 | ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC); | |
34dc7c2f BB |
564 | |
565 | *lp = l; | |
566 | return (0); | |
567 | } | |
568 | ||
569 | static int | |
570 | zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp) | |
571 | { | |
572 | ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); | |
573 | ||
d683ddbb | 574 | if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) { |
34dc7c2f | 575 | ASSERT3U(idx, <, |
d683ddbb | 576 | (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift)); |
34dc7c2f BB |
577 | *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx); |
578 | return (0); | |
579 | } else { | |
d683ddbb | 580 | return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl, |
34dc7c2f BB |
581 | idx, valp)); |
582 | } | |
583 | } | |
584 | ||
585 | static int | |
586 | zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx) | |
587 | { | |
588 | ASSERT(tx != NULL); | |
589 | ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); | |
590 | ||
d683ddbb | 591 | if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) { |
34dc7c2f BB |
592 | ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk; |
593 | return (0); | |
594 | } else { | |
d683ddbb | 595 | return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl, |
34dc7c2f BB |
596 | idx, blk, tx)); |
597 | } | |
598 | } | |
599 | ||
600 | static int | |
601 | zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp) | |
602 | { | |
d2a12f9e | 603 | uint64_t blk; |
34dc7c2f BB |
604 | |
605 | ASSERT(zap->zap_dbuf == NULL || | |
d683ddbb | 606 | zap_f_phys(zap) == zap->zap_dbuf->db_data); |
32c8c946 CC |
607 | |
608 | /* Reality check for corrupt zap objects (leaf or header). */ | |
609 | if ((zap_f_phys(zap)->zap_block_type != ZBT_LEAF && | |
610 | zap_f_phys(zap)->zap_block_type != ZBT_HEADER) || | |
611 | zap_f_phys(zap)->zap_magic != ZAP_MAGIC) { | |
612 | return (SET_ERROR(EIO)); | |
613 | } | |
d2a12f9e MA |
614 | |
615 | uint64_t idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift); | |
616 | int err = zap_idx_to_blk(zap, idx, &blk); | |
34dc7c2f BB |
617 | if (err != 0) |
618 | return (err); | |
619 | err = zap_get_leaf_byblk(zap, blk, tx, lt, lp); | |
620 | ||
d683ddbb JG |
621 | ASSERT(err || |
622 | ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) == | |
623 | zap_leaf_phys(*lp)->l_hdr.lh_prefix); | |
34dc7c2f BB |
624 | return (err); |
625 | } | |
626 | ||
627 | static int | |
8bea9815 | 628 | zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, |
dd66857d | 629 | const void *tag, dmu_tx_t *tx, zap_leaf_t **lp) |
34dc7c2f BB |
630 | { |
631 | zap_t *zap = zn->zn_zap; | |
632 | uint64_t hash = zn->zn_hash; | |
d2a12f9e | 633 | int err; |
d683ddbb | 634 | int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len; |
34dc7c2f | 635 | |
d683ddbb | 636 | ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift); |
34dc7c2f BB |
637 | ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); |
638 | ||
639 | ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==, | |
d683ddbb | 640 | zap_leaf_phys(l)->l_hdr.lh_prefix); |
34dc7c2f BB |
641 | |
642 | if (zap_tryupgradedir(zap, tx) == 0 || | |
d683ddbb | 643 | old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) { |
34dc7c2f BB |
644 | /* We failed to upgrade, or need to grow the pointer table */ |
645 | objset_t *os = zap->zap_objset; | |
646 | uint64_t object = zap->zap_object; | |
647 | ||
648 | zap_put_leaf(l); | |
8bea9815 | 649 | zap_unlockdir(zap, tag); |
34dc7c2f | 650 | err = zap_lockdir(os, object, tx, RW_WRITER, |
8bea9815 | 651 | FALSE, FALSE, tag, &zn->zn_zap); |
34dc7c2f | 652 | zap = zn->zn_zap; |
d2a12f9e | 653 | if (err != 0) |
34dc7c2f BB |
654 | return (err); |
655 | ASSERT(!zap->zap_ismicro); | |
656 | ||
657 | while (old_prefix_len == | |
d683ddbb | 658 | zap_f_phys(zap)->zap_ptrtbl.zt_shift) { |
34dc7c2f | 659 | err = zap_grow_ptrtbl(zap, tx); |
d2a12f9e | 660 | if (err != 0) |
34dc7c2f BB |
661 | return (err); |
662 | } | |
663 | ||
664 | err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l); | |
d2a12f9e | 665 | if (err != 0) |
34dc7c2f BB |
666 | return (err); |
667 | ||
d683ddbb | 668 | if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) { |
34dc7c2f BB |
669 | /* it split while our locks were down */ |
670 | *lp = l; | |
671 | return (0); | |
672 | } | |
673 | } | |
674 | ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); | |
d683ddbb | 675 | ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift); |
34dc7c2f | 676 | ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==, |
d683ddbb | 677 | zap_leaf_phys(l)->l_hdr.lh_prefix); |
34dc7c2f | 678 | |
d2a12f9e | 679 | int prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift - |
34dc7c2f | 680 | (old_prefix_len + 1); |
d2a12f9e MA |
681 | uint64_t sibling = |
682 | (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff; | |
34dc7c2f BB |
683 | |
684 | /* check for i/o errors before doing zap_leaf_split */ | |
d2a12f9e | 685 | for (int i = 0; i < (1ULL << prefix_diff); i++) { |
34dc7c2f | 686 | uint64_t blk; |
d2a12f9e MA |
687 | err = zap_idx_to_blk(zap, sibling + i, &blk); |
688 | if (err != 0) | |
34dc7c2f BB |
689 | return (err); |
690 | ASSERT3U(blk, ==, l->l_blkid); | |
691 | } | |
692 | ||
d2a12f9e | 693 | zap_leaf_t *nl = zap_create_leaf(zap, tx); |
34dc7c2f BB |
694 | zap_leaf_split(l, nl, zap->zap_normflags != 0); |
695 | ||
696 | /* set sibling pointers */ | |
d2a12f9e MA |
697 | for (int i = 0; i < (1ULL << prefix_diff); i++) { |
698 | err = zap_set_idx_to_blk(zap, sibling + i, nl->l_blkid, tx); | |
c99c9001 | 699 | ASSERT0(err); /* we checked for i/o errors above */ |
34dc7c2f BB |
700 | } |
701 | ||
6ca636a1 GN |
702 | ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_prefix_len, >, 0); |
703 | ||
d683ddbb | 704 | if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) { |
34dc7c2f BB |
705 | /* we want the sibling */ |
706 | zap_put_leaf(l); | |
707 | *lp = nl; | |
708 | } else { | |
709 | zap_put_leaf(nl); | |
710 | *lp = l; | |
711 | } | |
712 | ||
713 | return (0); | |
714 | } | |
715 | ||
716 | static void | |
8bea9815 | 717 | zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, |
dd66857d | 718 | const void *tag, dmu_tx_t *tx) |
34dc7c2f BB |
719 | { |
720 | zap_t *zap = zn->zn_zap; | |
d683ddbb JG |
721 | int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift; |
722 | int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift && | |
723 | zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER); | |
34dc7c2f BB |
724 | |
725 | zap_put_leaf(l); | |
726 | ||
d683ddbb | 727 | if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) { |
34dc7c2f BB |
728 | /* |
729 | * We are in the middle of growing the pointer table, or | |
730 | * this leaf will soon make us grow it. | |
731 | */ | |
732 | if (zap_tryupgradedir(zap, tx) == 0) { | |
733 | objset_t *os = zap->zap_objset; | |
734 | uint64_t zapobj = zap->zap_object; | |
735 | ||
8bea9815 | 736 | zap_unlockdir(zap, tag); |
d2a12f9e | 737 | int err = zap_lockdir(os, zapobj, tx, |
8bea9815 | 738 | RW_WRITER, FALSE, FALSE, tag, &zn->zn_zap); |
34dc7c2f | 739 | zap = zn->zn_zap; |
d2a12f9e | 740 | if (err != 0) |
34dc7c2f BB |
741 | return; |
742 | } | |
743 | ||
744 | /* could have finished growing while our locks were down */ | |
d683ddbb | 745 | if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift) |
34dc7c2f BB |
746 | (void) zap_grow_ptrtbl(zap, tx); |
747 | } | |
748 | } | |
749 | ||
34dc7c2f | 750 | static int |
428870ff | 751 | fzap_checkname(zap_name_t *zn) |
34dc7c2f | 752 | { |
428870ff | 753 | if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN) |
2e528b49 | 754 | return (SET_ERROR(ENAMETOOLONG)); |
428870ff BB |
755 | return (0); |
756 | } | |
34dc7c2f | 757 | |
428870ff BB |
758 | static int |
759 | fzap_checksize(uint64_t integer_size, uint64_t num_integers) | |
760 | { | |
34dc7c2f BB |
761 | /* Only integer sizes supported by C */ |
762 | switch (integer_size) { | |
763 | case 1: | |
764 | case 2: | |
765 | case 4: | |
766 | case 8: | |
767 | break; | |
768 | default: | |
2e528b49 | 769 | return (SET_ERROR(EINVAL)); |
34dc7c2f BB |
770 | } |
771 | ||
772 | if (integer_size * num_integers > ZAP_MAXVALUELEN) | |
ecb2b7dc | 773 | return (SET_ERROR(E2BIG)); |
34dc7c2f BB |
774 | |
775 | return (0); | |
776 | } | |
777 | ||
428870ff BB |
778 | static int |
779 | fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers) | |
780 | { | |
d2a12f9e MA |
781 | int err = fzap_checkname(zn); |
782 | if (err != 0) | |
428870ff BB |
783 | return (err); |
784 | return (fzap_checksize(integer_size, num_integers)); | |
785 | } | |
786 | ||
34dc7c2f BB |
787 | /* |
788 | * Routines for manipulating attributes. | |
789 | */ | |
790 | int | |
791 | fzap_lookup(zap_name_t *zn, | |
792 | uint64_t integer_size, uint64_t num_integers, void *buf, | |
793 | char *realname, int rn_len, boolean_t *ncp) | |
794 | { | |
795 | zap_leaf_t *l; | |
34dc7c2f BB |
796 | zap_entry_handle_t zeh; |
797 | ||
d2a12f9e MA |
798 | int err = fzap_checkname(zn); |
799 | if (err != 0) | |
34dc7c2f BB |
800 | return (err); |
801 | ||
802 | err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l); | |
803 | if (err != 0) | |
804 | return (err); | |
805 | err = zap_leaf_lookup(l, zn, &zeh); | |
806 | if (err == 0) { | |
428870ff BB |
807 | if ((err = fzap_checksize(integer_size, num_integers)) != 0) { |
808 | zap_put_leaf(l); | |
809 | return (err); | |
810 | } | |
811 | ||
34dc7c2f | 812 | err = zap_entry_read(&zeh, integer_size, num_integers, buf); |
428870ff | 813 | (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname); |
34dc7c2f BB |
814 | if (ncp) { |
815 | *ncp = zap_entry_normalization_conflict(&zeh, | |
816 | zn, NULL, zn->zn_zap); | |
817 | } | |
818 | } | |
819 | ||
820 | zap_put_leaf(l); | |
821 | return (err); | |
822 | } | |
823 | ||
824 | int | |
825 | fzap_add_cd(zap_name_t *zn, | |
826 | uint64_t integer_size, uint64_t num_integers, | |
dd66857d | 827 | const void *val, uint32_t cd, const void *tag, dmu_tx_t *tx) |
34dc7c2f BB |
828 | { |
829 | zap_leaf_t *l; | |
830 | int err; | |
831 | zap_entry_handle_t zeh; | |
832 | zap_t *zap = zn->zn_zap; | |
833 | ||
834 | ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); | |
835 | ASSERT(!zap->zap_ismicro); | |
428870ff | 836 | ASSERT(fzap_check(zn, integer_size, num_integers) == 0); |
34dc7c2f BB |
837 | |
838 | err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l); | |
839 | if (err != 0) | |
840 | return (err); | |
841 | retry: | |
842 | err = zap_leaf_lookup(l, zn, &zeh); | |
843 | if (err == 0) { | |
2e528b49 | 844 | err = SET_ERROR(EEXIST); |
34dc7c2f BB |
845 | goto out; |
846 | } | |
847 | if (err != ENOENT) | |
848 | goto out; | |
849 | ||
428870ff | 850 | err = zap_entry_create(l, zn, cd, |
34dc7c2f BB |
851 | integer_size, num_integers, val, &zeh); |
852 | ||
853 | if (err == 0) { | |
854 | zap_increment_num_entries(zap, 1, tx); | |
855 | } else if (err == EAGAIN) { | |
8bea9815 | 856 | err = zap_expand_leaf(zn, l, tag, tx, &l); |
34dc7c2f | 857 | zap = zn->zn_zap; /* zap_expand_leaf() may change zap */ |
599b8648 | 858 | if (err == 0) { |
34dc7c2f | 859 | goto retry; |
599b8648 CC |
860 | } else if (err == ENOSPC) { |
861 | /* | |
862 | * If we failed to expand the leaf, then bailout | |
863 | * as there is no point trying | |
864 | * zap_put_leaf_maybe_grow_ptrtbl(). | |
865 | */ | |
866 | return (err); | |
867 | } | |
34dc7c2f BB |
868 | } |
869 | ||
870 | out: | |
871 | if (zap != NULL) | |
8bea9815 | 872 | zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx); |
34dc7c2f BB |
873 | return (err); |
874 | } | |
875 | ||
876 | int | |
877 | fzap_add(zap_name_t *zn, | |
878 | uint64_t integer_size, uint64_t num_integers, | |
dd66857d | 879 | const void *val, const void *tag, dmu_tx_t *tx) |
34dc7c2f | 880 | { |
428870ff | 881 | int err = fzap_check(zn, integer_size, num_integers); |
34dc7c2f BB |
882 | if (err != 0) |
883 | return (err); | |
884 | ||
885 | return (fzap_add_cd(zn, integer_size, num_integers, | |
8bea9815 | 886 | val, ZAP_NEED_CD, tag, tx)); |
34dc7c2f BB |
887 | } |
888 | ||
889 | int | |
890 | fzap_update(zap_name_t *zn, | |
8bea9815 | 891 | int integer_size, uint64_t num_integers, const void *val, |
dd66857d | 892 | const void *tag, dmu_tx_t *tx) |
34dc7c2f BB |
893 | { |
894 | zap_leaf_t *l; | |
d2a12f9e MA |
895 | int err; |
896 | boolean_t create; | |
34dc7c2f BB |
897 | zap_entry_handle_t zeh; |
898 | zap_t *zap = zn->zn_zap; | |
899 | ||
900 | ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); | |
428870ff | 901 | err = fzap_check(zn, integer_size, num_integers); |
34dc7c2f BB |
902 | if (err != 0) |
903 | return (err); | |
904 | ||
905 | err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l); | |
906 | if (err != 0) | |
907 | return (err); | |
908 | retry: | |
909 | err = zap_leaf_lookup(l, zn, &zeh); | |
910 | create = (err == ENOENT); | |
911 | ASSERT(err == 0 || err == ENOENT); | |
912 | ||
913 | if (create) { | |
428870ff BB |
914 | err = zap_entry_create(l, zn, ZAP_NEED_CD, |
915 | integer_size, num_integers, val, &zeh); | |
34dc7c2f BB |
916 | if (err == 0) |
917 | zap_increment_num_entries(zap, 1, tx); | |
918 | } else { | |
919 | err = zap_entry_update(&zeh, integer_size, num_integers, val); | |
920 | } | |
921 | ||
922 | if (err == EAGAIN) { | |
8bea9815 | 923 | err = zap_expand_leaf(zn, l, tag, tx, &l); |
34dc7c2f BB |
924 | zap = zn->zn_zap; /* zap_expand_leaf() may change zap */ |
925 | if (err == 0) | |
926 | goto retry; | |
927 | } | |
928 | ||
929 | if (zap != NULL) | |
8bea9815 | 930 | zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx); |
34dc7c2f BB |
931 | return (err); |
932 | } | |
933 | ||
934 | int | |
935 | fzap_length(zap_name_t *zn, | |
936 | uint64_t *integer_size, uint64_t *num_integers) | |
937 | { | |
938 | zap_leaf_t *l; | |
939 | int err; | |
940 | zap_entry_handle_t zeh; | |
941 | ||
942 | err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l); | |
943 | if (err != 0) | |
944 | return (err); | |
945 | err = zap_leaf_lookup(l, zn, &zeh); | |
946 | if (err != 0) | |
947 | goto out; | |
948 | ||
4ef69de3 | 949 | if (integer_size != NULL) |
34dc7c2f | 950 | *integer_size = zeh.zeh_integer_size; |
4ef69de3 | 951 | if (num_integers != NULL) |
34dc7c2f BB |
952 | *num_integers = zeh.zeh_num_integers; |
953 | out: | |
954 | zap_put_leaf(l); | |
955 | return (err); | |
956 | } | |
957 | ||
958 | int | |
959 | fzap_remove(zap_name_t *zn, dmu_tx_t *tx) | |
960 | { | |
961 | zap_leaf_t *l; | |
962 | int err; | |
963 | zap_entry_handle_t zeh; | |
964 | ||
965 | err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l); | |
966 | if (err != 0) | |
967 | return (err); | |
968 | err = zap_leaf_lookup(l, zn, &zeh); | |
969 | if (err == 0) { | |
970 | zap_entry_remove(&zeh); | |
971 | zap_increment_num_entries(zn->zn_zap, -1, tx); | |
972 | } | |
973 | zap_put_leaf(l); | |
974 | return (err); | |
975 | } | |
976 | ||
428870ff BB |
977 | void |
978 | fzap_prefetch(zap_name_t *zn) | |
979 | { | |
d2a12f9e | 980 | uint64_t blk; |
428870ff | 981 | zap_t *zap = zn->zn_zap; |
428870ff | 982 | |
d2a12f9e | 983 | uint64_t idx = ZAP_HASH_IDX(zn->zn_hash, |
d683ddbb | 984 | zap_f_phys(zap)->zap_ptrtbl.zt_shift); |
428870ff BB |
985 | if (zap_idx_to_blk(zap, idx, &blk) != 0) |
986 | return; | |
d2a12f9e | 987 | int bs = FZAP_BLOCK_SHIFT(zap); |
fcff0f35 PD |
988 | dmu_prefetch(zap->zap_objset, zap->zap_object, 0, blk << bs, 1 << bs, |
989 | ZIO_PRIORITY_SYNC_READ); | |
428870ff BB |
990 | } |
991 | ||
b128c09f BB |
992 | /* |
993 | * Helper functions for consumers. | |
994 | */ | |
995 | ||
9ae529ec CS |
996 | uint64_t |
997 | zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj, | |
998 | const char *name, dmu_tx_t *tx) | |
50c957f7 NB |
999 | { |
1000 | return (zap_create_link_dnsize(os, ot, parent_obj, name, 0, tx)); | |
1001 | } | |
1002 | ||
1003 | uint64_t | |
1004 | zap_create_link_dnsize(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj, | |
1005 | const char *name, int dnodesize, dmu_tx_t *tx) | |
9ae529ec CS |
1006 | { |
1007 | uint64_t new_obj; | |
1008 | ||
d2a12f9e MA |
1009 | new_obj = zap_create_dnsize(os, ot, DMU_OT_NONE, 0, dnodesize, tx); |
1010 | VERIFY(new_obj != 0); | |
e0ab3ab5 JS |
1011 | VERIFY0(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj, |
1012 | tx)); | |
9ae529ec CS |
1013 | |
1014 | return (new_obj); | |
1015 | } | |
1016 | ||
34dc7c2f BB |
1017 | int |
1018 | zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask, | |
1019 | char *name) | |
1020 | { | |
1021 | zap_cursor_t zc; | |
34dc7c2f BB |
1022 | int err; |
1023 | ||
1024 | if (mask == 0) | |
1025 | mask = -1ULL; | |
1026 | ||
d2a12f9e | 1027 | zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP); |
34dc7c2f BB |
1028 | for (zap_cursor_init(&zc, os, zapobj); |
1029 | (err = zap_cursor_retrieve(&zc, za)) == 0; | |
1030 | zap_cursor_advance(&zc)) { | |
1031 | if ((za->za_first_integer & mask) == (value & mask)) { | |
c9e319fa | 1032 | (void) strlcpy(name, za->za_name, MAXNAMELEN); |
34dc7c2f BB |
1033 | break; |
1034 | } | |
1035 | } | |
1036 | zap_cursor_fini(&zc); | |
d2a12f9e | 1037 | kmem_free(za, sizeof (*za)); |
34dc7c2f BB |
1038 | return (err); |
1039 | } | |
1040 | ||
b128c09f BB |
1041 | int |
1042 | zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx) | |
1043 | { | |
1044 | zap_cursor_t zc; | |
d2a12f9e | 1045 | int err = 0; |
b128c09f | 1046 | |
d2a12f9e | 1047 | zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP); |
b128c09f | 1048 | for (zap_cursor_init(&zc, os, fromobj); |
d2a12f9e | 1049 | zap_cursor_retrieve(&zc, za) == 0; |
b128c09f | 1050 | (void) zap_cursor_advance(&zc)) { |
d2a12f9e | 1051 | if (za->za_integer_length != 8 || za->za_num_integers != 1) { |
3a84951d WA |
1052 | err = SET_ERROR(EINVAL); |
1053 | break; | |
1054 | } | |
d2a12f9e MA |
1055 | err = zap_add(os, intoobj, za->za_name, |
1056 | 8, 1, &za->za_first_integer, tx); | |
1057 | if (err != 0) | |
3a84951d | 1058 | break; |
b128c09f BB |
1059 | } |
1060 | zap_cursor_fini(&zc); | |
d2a12f9e | 1061 | kmem_free(za, sizeof (*za)); |
3a84951d | 1062 | return (err); |
b128c09f BB |
1063 | } |
1064 | ||
428870ff BB |
1065 | int |
1066 | zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj, | |
1067 | uint64_t value, dmu_tx_t *tx) | |
1068 | { | |
1069 | zap_cursor_t zc; | |
d2a12f9e | 1070 | int err = 0; |
428870ff | 1071 | |
d2a12f9e | 1072 | zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP); |
428870ff | 1073 | for (zap_cursor_init(&zc, os, fromobj); |
d2a12f9e | 1074 | zap_cursor_retrieve(&zc, za) == 0; |
428870ff | 1075 | (void) zap_cursor_advance(&zc)) { |
d2a12f9e | 1076 | if (za->za_integer_length != 8 || za->za_num_integers != 1) { |
3a84951d WA |
1077 | err = SET_ERROR(EINVAL); |
1078 | break; | |
1079 | } | |
d2a12f9e | 1080 | err = zap_add(os, intoobj, za->za_name, |
428870ff | 1081 | 8, 1, &value, tx); |
d4a72f23 | 1082 | if (err != 0) |
3a84951d | 1083 | break; |
428870ff BB |
1084 | } |
1085 | zap_cursor_fini(&zc); | |
d2a12f9e | 1086 | kmem_free(za, sizeof (*za)); |
3a84951d | 1087 | return (err); |
428870ff BB |
1088 | } |
1089 | ||
1090 | int | |
1091 | zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj, | |
1092 | dmu_tx_t *tx) | |
1093 | { | |
1094 | zap_cursor_t zc; | |
d2a12f9e | 1095 | int err = 0; |
428870ff | 1096 | |
d2a12f9e | 1097 | zap_attribute_t *za = kmem_alloc(sizeof (*za), KM_SLEEP); |
428870ff | 1098 | for (zap_cursor_init(&zc, os, fromobj); |
d2a12f9e | 1099 | zap_cursor_retrieve(&zc, za) == 0; |
428870ff BB |
1100 | (void) zap_cursor_advance(&zc)) { |
1101 | uint64_t delta = 0; | |
1102 | ||
d2a12f9e | 1103 | if (za->za_integer_length != 8 || za->za_num_integers != 1) { |
3a84951d WA |
1104 | err = SET_ERROR(EINVAL); |
1105 | break; | |
1106 | } | |
428870ff | 1107 | |
d2a12f9e | 1108 | err = zap_lookup(os, intoobj, za->za_name, 8, 1, &delta); |
428870ff | 1109 | if (err != 0 && err != ENOENT) |
3a84951d | 1110 | break; |
d2a12f9e MA |
1111 | delta += za->za_first_integer; |
1112 | err = zap_update(os, intoobj, za->za_name, 8, 1, &delta, tx); | |
1113 | if (err != 0) | |
3a84951d | 1114 | break; |
428870ff BB |
1115 | } |
1116 | zap_cursor_fini(&zc); | |
d2a12f9e | 1117 | kmem_free(za, sizeof (*za)); |
3a84951d | 1118 | return (err); |
428870ff BB |
1119 | } |
1120 | ||
b128c09f BB |
1121 | int |
1122 | zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx) | |
1123 | { | |
1124 | char name[20]; | |
1125 | ||
1126 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); | |
1127 | return (zap_add(os, obj, name, 8, 1, &value, tx)); | |
1128 | } | |
1129 | ||
1130 | int | |
1131 | zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx) | |
1132 | { | |
1133 | char name[20]; | |
1134 | ||
1135 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); | |
1136 | return (zap_remove(os, obj, name, tx)); | |
1137 | } | |
1138 | ||
1139 | int | |
1140 | zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value) | |
1141 | { | |
1142 | char name[20]; | |
1143 | ||
1144 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); | |
1145 | return (zap_lookup(os, obj, name, 8, 1, &value)); | |
1146 | } | |
34dc7c2f | 1147 | |
428870ff BB |
1148 | int |
1149 | zap_add_int_key(objset_t *os, uint64_t obj, | |
1150 | uint64_t key, uint64_t value, dmu_tx_t *tx) | |
1151 | { | |
1152 | char name[20]; | |
1153 | ||
1154 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); | |
1155 | return (zap_add(os, obj, name, 8, 1, &value, tx)); | |
1156 | } | |
1157 | ||
753c3839 MA |
1158 | int |
1159 | zap_update_int_key(objset_t *os, uint64_t obj, | |
1160 | uint64_t key, uint64_t value, dmu_tx_t *tx) | |
1161 | { | |
1162 | char name[20]; | |
1163 | ||
1164 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); | |
1165 | return (zap_update(os, obj, name, 8, 1, &value, tx)); | |
1166 | } | |
1167 | ||
428870ff BB |
1168 | int |
1169 | zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep) | |
1170 | { | |
1171 | char name[20]; | |
1172 | ||
1173 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); | |
1174 | return (zap_lookup(os, obj, name, 8, 1, valuep)); | |
1175 | } | |
1176 | ||
1177 | int | |
1178 | zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta, | |
1179 | dmu_tx_t *tx) | |
1180 | { | |
1181 | uint64_t value = 0; | |
428870ff BB |
1182 | |
1183 | if (delta == 0) | |
1184 | return (0); | |
1185 | ||
d2a12f9e | 1186 | int err = zap_lookup(os, obj, name, 8, 1, &value); |
428870ff BB |
1187 | if (err != 0 && err != ENOENT) |
1188 | return (err); | |
1189 | value += delta; | |
1190 | if (value == 0) | |
1191 | err = zap_remove(os, obj, name, tx); | |
1192 | else | |
1193 | err = zap_update(os, obj, name, 8, 1, &value, tx); | |
1194 | return (err); | |
1195 | } | |
1196 | ||
1197 | int | |
1198 | zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta, | |
1199 | dmu_tx_t *tx) | |
1200 | { | |
1201 | char name[20]; | |
1202 | ||
1203 | (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); | |
1204 | return (zap_increment(os, obj, name, delta, tx)); | |
1205 | } | |
1206 | ||
34dc7c2f BB |
1207 | /* |
1208 | * Routines for iterating over the attributes. | |
1209 | */ | |
1210 | ||
1211 | int | |
1212 | fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za) | |
1213 | { | |
1214 | int err = ENOENT; | |
1215 | zap_entry_handle_t zeh; | |
1216 | zap_leaf_t *l; | |
1217 | ||
1218 | /* retrieve the next entry at or after zc_hash/zc_cd */ | |
1219 | /* if no entry, return ENOENT */ | |
1220 | ||
d9b4bf06 MA |
1221 | /* |
1222 | * If we are reading from the beginning, we're almost certain to | |
1223 | * iterate over the entire ZAP object. If there are multiple leaf | |
1224 | * blocks (freeblk > 2), prefetch the whole object (up to | |
1225 | * dmu_prefetch_max bytes), so that we read the leaf blocks | |
1226 | * concurrently. (Unless noprefetch was requested via | |
1227 | * zap_cursor_init_noprefetch()). | |
1228 | */ | |
1229 | if (zc->zc_hash == 0 && zap_iterate_prefetch && | |
1230 | zc->zc_prefetch && zap_f_phys(zap)->zap_freeblk > 2) { | |
1231 | dmu_prefetch(zc->zc_objset, zc->zc_zapobj, 0, 0, | |
1232 | zap_f_phys(zap)->zap_freeblk << FZAP_BLOCK_SHIFT(zap), | |
1233 | ZIO_PRIORITY_ASYNC_READ); | |
1234 | } | |
1235 | ||
34dc7c2f BB |
1236 | if (zc->zc_leaf && |
1237 | (ZAP_HASH_IDX(zc->zc_hash, | |
d683ddbb JG |
1238 | zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) != |
1239 | zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) { | |
34dc7c2f BB |
1240 | rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); |
1241 | zap_put_leaf(zc->zc_leaf); | |
1242 | zc->zc_leaf = NULL; | |
1243 | } | |
1244 | ||
1245 | again: | |
1246 | if (zc->zc_leaf == NULL) { | |
1247 | err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER, | |
1248 | &zc->zc_leaf); | |
1249 | if (err != 0) | |
1250 | return (err); | |
1251 | } else { | |
1252 | rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); | |
1253 | } | |
1254 | l = zc->zc_leaf; | |
1255 | ||
1256 | err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh); | |
1257 | ||
1258 | if (err == ENOENT) { | |
6ca636a1 | 1259 | if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0) { |
34dc7c2f | 1260 | zc->zc_hash = -1ULL; |
6ca636a1 | 1261 | zc->zc_cd = 0; |
34dc7c2f | 1262 | } else { |
6ca636a1 GN |
1263 | uint64_t nocare = (1ULL << |
1264 | (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1; | |
1265 | ||
1266 | zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1; | |
1267 | zc->zc_cd = 0; | |
1268 | ||
1269 | if (zc->zc_hash == 0) { | |
1270 | zc->zc_hash = -1ULL; | |
1271 | } else { | |
1272 | zap_put_leaf(zc->zc_leaf); | |
1273 | zc->zc_leaf = NULL; | |
1274 | goto again; | |
1275 | } | |
34dc7c2f BB |
1276 | } |
1277 | } | |
1278 | ||
1279 | if (err == 0) { | |
1280 | zc->zc_hash = zeh.zeh_hash; | |
1281 | zc->zc_cd = zeh.zeh_cd; | |
1282 | za->za_integer_length = zeh.zeh_integer_size; | |
1283 | za->za_num_integers = zeh.zeh_num_integers; | |
1284 | if (zeh.zeh_num_integers == 0) { | |
1285 | za->za_first_integer = 0; | |
1286 | } else { | |
1287 | err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer); | |
1288 | ASSERT(err == 0 || err == EOVERFLOW); | |
1289 | } | |
428870ff | 1290 | err = zap_entry_read_name(zap, &zeh, |
34dc7c2f BB |
1291 | sizeof (za->za_name), za->za_name); |
1292 | ASSERT(err == 0); | |
1293 | ||
1294 | za->za_normalization_conflict = | |
1295 | zap_entry_normalization_conflict(&zeh, | |
1296 | NULL, za->za_name, zap); | |
1297 | } | |
1298 | rw_exit(&zc->zc_leaf->l_rwlock); | |
1299 | return (err); | |
1300 | } | |
1301 | ||
34dc7c2f BB |
1302 | static void |
1303 | zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs) | |
1304 | { | |
34dc7c2f BB |
1305 | uint64_t lastblk = 0; |
1306 | ||
1307 | /* | |
1308 | * NB: if a leaf has more pointers than an entire ptrtbl block | |
1309 | * can hold, then it'll be accounted for more than once, since | |
1310 | * we won't have lastblk. | |
1311 | */ | |
d2a12f9e | 1312 | for (int i = 0; i < len; i++) { |
34dc7c2f BB |
1313 | zap_leaf_t *l; |
1314 | ||
1315 | if (tbl[i] == lastblk) | |
1316 | continue; | |
1317 | lastblk = tbl[i]; | |
1318 | ||
d2a12f9e | 1319 | int err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l); |
34dc7c2f BB |
1320 | if (err == 0) { |
1321 | zap_leaf_stats(zap, l, zs); | |
1322 | zap_put_leaf(l); | |
1323 | } | |
1324 | } | |
1325 | } | |
1326 | ||
1327 | void | |
1328 | fzap_get_stats(zap_t *zap, zap_stats_t *zs) | |
1329 | { | |
1330 | int bs = FZAP_BLOCK_SHIFT(zap); | |
1331 | zs->zs_blocksize = 1ULL << bs; | |
1332 | ||
1333 | /* | |
1334 | * Set zap_phys_t fields | |
1335 | */ | |
d683ddbb JG |
1336 | zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs; |
1337 | zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries; | |
1338 | zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk; | |
1339 | zs->zs_block_type = zap_f_phys(zap)->zap_block_type; | |
1340 | zs->zs_magic = zap_f_phys(zap)->zap_magic; | |
1341 | zs->zs_salt = zap_f_phys(zap)->zap_salt; | |
34dc7c2f BB |
1342 | |
1343 | /* | |
1344 | * Set zap_ptrtbl fields | |
1345 | */ | |
d683ddbb JG |
1346 | zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift; |
1347 | zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk; | |
34dc7c2f | 1348 | zs->zs_ptrtbl_blks_copied = |
d683ddbb JG |
1349 | zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied; |
1350 | zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk; | |
1351 | zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks; | |
1352 | zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift; | |
34dc7c2f | 1353 | |
d683ddbb | 1354 | if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) { |
34dc7c2f BB |
1355 | /* the ptrtbl is entirely in the header block. */ |
1356 | zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), | |
1357 | 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs); | |
1358 | } else { | |
fcff0f35 | 1359 | dmu_prefetch(zap->zap_objset, zap->zap_object, 0, |
d683ddbb | 1360 | zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs, |
fcff0f35 PD |
1361 | zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs, |
1362 | ZIO_PRIORITY_SYNC_READ); | |
34dc7c2f | 1363 | |
d2a12f9e | 1364 | for (int b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks; |
34dc7c2f BB |
1365 | b++) { |
1366 | dmu_buf_t *db; | |
1367 | int err; | |
1368 | ||
1369 | err = dmu_buf_hold(zap->zap_objset, zap->zap_object, | |
d683ddbb | 1370 | (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs, |
428870ff | 1371 | FTAG, &db, DMU_READ_NO_PREFETCH); |
34dc7c2f BB |
1372 | if (err == 0) { |
1373 | zap_stats_ptrtbl(zap, db->db_data, | |
1374 | 1<<(bs-3), zs); | |
1375 | dmu_buf_rele(db, FTAG); | |
1376 | } | |
1377 | } | |
1378 | } | |
1379 | } | |
d9b4bf06 | 1380 | |
7ada752a | 1381 | /* CSTYLED */ |
03fdcb9a | 1382 | ZFS_MODULE_PARAM(zfs, , zap_iterate_prefetch, INT, ZMOD_RW, |
d9b4bf06 | 1383 | "When iterating ZAP object, prefetch it"); |