]>
Commit | Line | Data |
---|---|---|
da536844 MA |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * This file and its contents are supplied under the terms of the | |
5 | * Common Development and Distribution License ("CDDL"), version 1.0. | |
6 | * You may only use this file in accordance with the terms of version | |
7 | * 1.0 of the CDDL. | |
8 | * | |
9 | * A full copy of the text of the CDDL should have accompanied this | |
10 | * source. A copy of the CDDL is also available via the Internet at | |
11 | * http://www.illumos.org/license/CDDL. | |
12 | * | |
13 | * CDDL HEADER END | |
14 | */ | |
9b7b9cd3 | 15 | |
da536844 | 16 | /* |
30af21b0 | 17 | * Copyright (c) 2013, 2018 by Delphix. All rights reserved. |
9b7b9cd3 | 18 | * Copyright 2017 Nexenta Systems, Inc. |
a73f361f | 19 | * Copyright 2019, 2020 by Christian Schwarz. All rights reserved. |
da536844 MA |
20 | */ |
21 | ||
22 | #include <sys/zfs_context.h> | |
23 | #include <sys/dsl_dataset.h> | |
24 | #include <sys/dsl_dir.h> | |
25 | #include <sys/dsl_prop.h> | |
26 | #include <sys/dsl_synctask.h> | |
30af21b0 | 27 | #include <sys/dsl_destroy.h> |
da536844 MA |
28 | #include <sys/dmu_impl.h> |
29 | #include <sys/dmu_tx.h> | |
30 | #include <sys/arc.h> | |
31 | #include <sys/zap.h> | |
32 | #include <sys/zfeature.h> | |
33 | #include <sys/spa.h> | |
34 | #include <sys/dsl_bookmark.h> | |
35 | #include <zfs_namecheck.h> | |
30af21b0 | 36 | #include <sys/dmu_send.h> |
bee9cfb8 | 37 | #include <sys/dbuf.h> |
da536844 MA |
38 | |
39 | static int | |
40 | dsl_bookmark_hold_ds(dsl_pool_t *dp, const char *fullname, | |
dd66857d | 41 | dsl_dataset_t **dsp, const void *tag, char **shortnamep) |
da536844 | 42 | { |
eca7b760 | 43 | char buf[ZFS_MAX_DATASET_NAME_LEN]; |
da536844 MA |
44 | char *hashp; |
45 | ||
eca7b760 | 46 | if (strlen(fullname) >= ZFS_MAX_DATASET_NAME_LEN) |
da536844 MA |
47 | return (SET_ERROR(ENAMETOOLONG)); |
48 | hashp = strchr(fullname, '#'); | |
49 | if (hashp == NULL) | |
50 | return (SET_ERROR(EINVAL)); | |
51 | ||
52 | *shortnamep = hashp + 1; | |
53 | if (zfs_component_namecheck(*shortnamep, NULL, NULL)) | |
54 | return (SET_ERROR(EINVAL)); | |
55 | (void) strlcpy(buf, fullname, hashp - fullname + 1); | |
56 | return (dsl_dataset_hold(dp, buf, tag, dsp)); | |
57 | } | |
58 | ||
59 | /* | |
a73f361f CS |
60 | * When reading BOOKMARK_V1 bookmarks, the BOOKMARK_V2 fields are guaranteed |
61 | * to be zeroed. | |
62 | * | |
da536844 | 63 | * Returns ESRCH if bookmark is not found. |
30af21b0 PD |
64 | * Note, we need to use the ZAP rather than the AVL to look up bookmarks |
65 | * by name, because only the ZAP honors the casesensitivity setting. | |
da536844 | 66 | */ |
30af21b0 PD |
67 | int |
68 | dsl_bookmark_lookup_impl(dsl_dataset_t *ds, const char *shortname, | |
da536844 MA |
69 | zfs_bookmark_phys_t *bmark_phys) |
70 | { | |
71 | objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
30af21b0 | 72 | uint64_t bmark_zapobj = ds->ds_bookmarks_obj; |
9b7b9cd3 | 73 | matchtype_t mt = 0; |
da536844 MA |
74 | int err; |
75 | ||
76 | if (bmark_zapobj == 0) | |
77 | return (SET_ERROR(ESRCH)); | |
78 | ||
d683ddbb | 79 | if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) |
9b7b9cd3 | 80 | mt = MT_NORMALIZE; |
da536844 | 81 | |
579ce7c5 TC |
82 | /* |
83 | * Zero out the bookmark in case the one stored on disk | |
84 | * is in an older, shorter format. | |
85 | */ | |
861166b0 | 86 | memset(bmark_phys, 0, sizeof (*bmark_phys)); |
579ce7c5 | 87 | |
da536844 | 88 | err = zap_lookup_norm(mos, bmark_zapobj, shortname, sizeof (uint64_t), |
30af21b0 PD |
89 | sizeof (*bmark_phys) / sizeof (uint64_t), bmark_phys, mt, NULL, 0, |
90 | NULL); | |
da536844 | 91 | |
28caa74b | 92 | return (err == ENOENT ? SET_ERROR(ESRCH) : err); |
da536844 MA |
93 | } |
94 | ||
95 | /* | |
e1cfd73f | 96 | * If later_ds is non-NULL, this will return EXDEV if the specified bookmark |
30af21b0 PD |
97 | * does not represents an earlier point in later_ds's timeline. However, |
98 | * bmp will still be filled in if we return EXDEV. | |
da536844 MA |
99 | * |
100 | * Returns ENOENT if the dataset containing the bookmark does not exist. | |
101 | * Returns ESRCH if the dataset exists but the bookmark was not found in it. | |
102 | */ | |
103 | int | |
104 | dsl_bookmark_lookup(dsl_pool_t *dp, const char *fullname, | |
105 | dsl_dataset_t *later_ds, zfs_bookmark_phys_t *bmp) | |
106 | { | |
107 | char *shortname; | |
108 | dsl_dataset_t *ds; | |
109 | int error; | |
110 | ||
111 | error = dsl_bookmark_hold_ds(dp, fullname, &ds, FTAG, &shortname); | |
112 | if (error != 0) | |
113 | return (error); | |
114 | ||
30af21b0 | 115 | error = dsl_bookmark_lookup_impl(ds, shortname, bmp); |
da536844 MA |
116 | if (error == 0 && later_ds != NULL) { |
117 | if (!dsl_dataset_is_before(later_ds, ds, bmp->zbm_creation_txg)) | |
118 | error = SET_ERROR(EXDEV); | |
119 | } | |
120 | dsl_dataset_rele(ds, FTAG); | |
121 | return (error); | |
122 | } | |
123 | ||
a73f361f CS |
124 | /* |
125 | * Validates that | |
126 | * - bmark is a full dataset path of a bookmark (bookmark_namecheck) | |
dd4bc569 | 127 | * - source is a full path of a snapshot or bookmark |
a73f361f CS |
128 | * ({bookmark,snapshot}_namecheck) |
129 | * | |
130 | * Returns 0 if valid, -1 otherwise. | |
131 | */ | |
132 | static int | |
133 | dsl_bookmark_create_nvl_validate_pair(const char *bmark, const char *source) | |
134 | { | |
135 | if (bookmark_namecheck(bmark, NULL, NULL) != 0) | |
136 | return (-1); | |
137 | ||
138 | int is_bmark, is_snap; | |
139 | is_bmark = bookmark_namecheck(source, NULL, NULL) == 0; | |
140 | is_snap = snapshot_namecheck(source, NULL, NULL) == 0; | |
141 | if (!is_bmark && !is_snap) | |
142 | return (-1); | |
143 | ||
144 | return (0); | |
145 | } | |
146 | ||
147 | /* | |
148 | * Check that the given nvlist corresponds to the following schema: | |
149 | * { newbookmark -> source, ... } | |
150 | * where | |
151 | * - each pair passes dsl_bookmark_create_nvl_validate_pair | |
152 | * - all newbookmarks are in the same pool | |
153 | * - all newbookmarks have unique names | |
154 | * | |
155 | * Note that this function is only validates above schema. Callers must ensure | |
156 | * that the bookmarks can be created, e.g. that sources exist. | |
157 | * | |
158 | * Returns 0 if the nvlist adheres to above schema. | |
159 | * Returns -1 if it doesn't. | |
160 | */ | |
161 | int | |
162 | dsl_bookmark_create_nvl_validate(nvlist_t *bmarks) | |
163 | { | |
d1807f16 | 164 | const char *first = NULL; |
7db823bd | 165 | size_t first_len = 0; |
a73f361f | 166 | |
a73f361f CS |
167 | for (nvpair_t *pair = nvlist_next_nvpair(bmarks, NULL); |
168 | pair != NULL; pair = nvlist_next_nvpair(bmarks, pair)) { | |
169 | ||
d1807f16 RY |
170 | const char *bmark = nvpair_name(pair); |
171 | const char *source; | |
a73f361f CS |
172 | |
173 | /* list structure: values must be snapshots XOR bookmarks */ | |
174 | if (nvpair_value_string(pair, &source) != 0) | |
175 | return (-1); | |
176 | if (dsl_bookmark_create_nvl_validate_pair(bmark, source) != 0) | |
177 | return (-1); | |
178 | ||
179 | /* same pool check */ | |
180 | if (first == NULL) { | |
d1807f16 | 181 | const char *cp = strpbrk(bmark, "/#"); |
a73f361f CS |
182 | if (cp == NULL) |
183 | return (-1); | |
184 | first = bmark; | |
185 | first_len = cp - bmark; | |
186 | } | |
187 | if (strncmp(first, bmark, first_len) != 0) | |
188 | return (-1); | |
189 | switch (*(bmark + first_len)) { | |
190 | case '/': /* fallthrough */ | |
191 | case '#': | |
192 | break; | |
193 | default: | |
194 | return (-1); | |
195 | } | |
196 | ||
197 | /* unique newbookmark names; todo: O(n^2) */ | |
198 | for (nvpair_t *pair2 = nvlist_next_nvpair(bmarks, pair); | |
199 | pair2 != NULL; pair2 = nvlist_next_nvpair(bmarks, pair2)) { | |
200 | if (strcmp(nvpair_name(pair), nvpair_name(pair2)) == 0) | |
201 | return (-1); | |
202 | } | |
203 | ||
204 | } | |
205 | return (0); | |
206 | } | |
207 | ||
a73f361f CS |
208 | /* |
209 | * expects that newbm and source have been validated using | |
210 | * dsl_bookmark_create_nvl_validate_pair | |
211 | */ | |
da536844 | 212 | static int |
a73f361f CS |
213 | dsl_bookmark_create_check_impl(dsl_pool_t *dp, |
214 | const char *newbm, const char *source) | |
da536844 | 215 | { |
a73f361f CS |
216 | ASSERT0(dsl_bookmark_create_nvl_validate_pair(newbm, source)); |
217 | /* defer source namecheck until we know it's a snapshot or bookmark */ | |
da536844 | 218 | |
a73f361f CS |
219 | int error; |
220 | dsl_dataset_t *newbm_ds; | |
221 | char *newbm_short; | |
222 | zfs_bookmark_phys_t bmark_phys; | |
da536844 | 223 | |
a73f361f | 224 | error = dsl_bookmark_hold_ds(dp, newbm, &newbm_ds, FTAG, &newbm_short); |
da536844 MA |
225 | if (error != 0) |
226 | return (error); | |
227 | ||
a73f361f CS |
228 | /* Verify that the new bookmark does not already exist */ |
229 | error = dsl_bookmark_lookup_impl(newbm_ds, newbm_short, &bmark_phys); | |
230 | switch (error) { | |
231 | case ESRCH: | |
232 | /* happy path: new bmark doesn't exist, proceed after switch */ | |
a73f361f CS |
233 | break; |
234 | case 0: | |
235 | error = SET_ERROR(EEXIST); | |
236 | goto eholdnewbmds; | |
237 | default: | |
bf169e9f | 238 | /* dsl_bookmark_lookup_impl already did SET_ERROR */ |
a73f361f | 239 | goto eholdnewbmds; |
da536844 MA |
240 | } |
241 | ||
a73f361f CS |
242 | /* error is retval of the following if-cascade */ |
243 | if (strchr(source, '@') != NULL) { | |
244 | dsl_dataset_t *source_snap_ds; | |
245 | ASSERT3S(snapshot_namecheck(source, NULL, NULL), ==, 0); | |
246 | error = dsl_dataset_hold(dp, source, FTAG, &source_snap_ds); | |
247 | if (error == 0) { | |
248 | VERIFY(source_snap_ds->ds_is_snapshot); | |
249 | /* | |
250 | * Verify that source snapshot is an earlier point in | |
251 | * newbm_ds's timeline (source may be newbm_ds's origin) | |
252 | */ | |
253 | if (!dsl_dataset_is_before(newbm_ds, source_snap_ds, 0)) | |
254 | error = SET_ERROR( | |
255 | ZFS_ERR_BOOKMARK_SOURCE_NOT_ANCESTOR); | |
256 | dsl_dataset_rele(source_snap_ds, FTAG); | |
257 | } | |
258 | } else if (strchr(source, '#') != NULL) { | |
259 | zfs_bookmark_phys_t source_phys; | |
260 | ASSERT3S(bookmark_namecheck(source, NULL, NULL), ==, 0); | |
261 | /* | |
262 | * Source must exists and be an earlier point in newbm_ds's | |
263 | * timeline (newbm_ds's origin may be a snap of source's ds) | |
264 | */ | |
265 | error = dsl_bookmark_lookup(dp, source, newbm_ds, &source_phys); | |
266 | switch (error) { | |
267 | case 0: | |
268 | break; /* happy path */ | |
269 | case EXDEV: | |
270 | error = SET_ERROR(ZFS_ERR_BOOKMARK_SOURCE_NOT_ANCESTOR); | |
271 | break; | |
272 | default: | |
bf169e9f | 273 | /* dsl_bookmark_lookup already did SET_ERROR */ |
a73f361f CS |
274 | break; |
275 | } | |
276 | } else { | |
277 | /* | |
278 | * dsl_bookmark_create_nvl_validate validates that source is | |
279 | * either snapshot or bookmark | |
280 | */ | |
281 | panic("unreachable code: %s", source); | |
282 | } | |
283 | ||
284 | eholdnewbmds: | |
285 | dsl_dataset_rele(newbm_ds, FTAG); | |
da536844 MA |
286 | return (error); |
287 | } | |
288 | ||
948f0c44 | 289 | int |
da536844 MA |
290 | dsl_bookmark_create_check(void *arg, dmu_tx_t *tx) |
291 | { | |
292 | dsl_bookmark_create_arg_t *dbca = arg; | |
a73f361f CS |
293 | int rv = 0; |
294 | int schema_err = 0; | |
20ea8540 CS |
295 | ASSERT3P(dbca, !=, NULL); |
296 | ASSERT3P(dbca->dbca_bmarks, !=, NULL); | |
a73f361f | 297 | /* dbca->dbca_errors is allowed to be NULL */ |
20ea8540 | 298 | |
da536844 | 299 | dsl_pool_t *dp = dmu_tx_pool(tx); |
da536844 MA |
300 | |
301 | if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARKS)) | |
302 | return (SET_ERROR(ENOTSUP)); | |
303 | ||
a73f361f CS |
304 | if (dsl_bookmark_create_nvl_validate(dbca->dbca_bmarks) != 0) |
305 | rv = schema_err = SET_ERROR(EINVAL); | |
306 | ||
1c27024e | 307 | for (nvpair_t *pair = nvlist_next_nvpair(dbca->dbca_bmarks, NULL); |
da536844 | 308 | pair != NULL; pair = nvlist_next_nvpair(dbca->dbca_bmarks, pair)) { |
d1807f16 | 309 | const char *new = nvpair_name(pair); |
da536844 | 310 | |
a73f361f | 311 | int error = schema_err; |
da536844 | 312 | if (error == 0) { |
d1807f16 | 313 | const char *source = fnvpair_value_string(pair); |
a73f361f CS |
314 | error = dsl_bookmark_create_check_impl(dp, new, source); |
315 | if (error != 0) | |
316 | error = SET_ERROR(error); | |
da536844 | 317 | } |
a73f361f | 318 | |
da536844 | 319 | if (error != 0) { |
da536844 | 320 | rv = error; |
20ea8540 CS |
321 | if (dbca->dbca_errors != NULL) |
322 | fnvlist_add_int32(dbca->dbca_errors, | |
a73f361f | 323 | new, error); |
da536844 MA |
324 | } |
325 | } | |
326 | ||
327 | return (rv); | |
328 | } | |
329 | ||
30af21b0 PD |
330 | static dsl_bookmark_node_t * |
331 | dsl_bookmark_node_alloc(char *shortname) | |
332 | { | |
333 | dsl_bookmark_node_t *dbn = kmem_alloc(sizeof (*dbn), KM_SLEEP); | |
334 | dbn->dbn_name = spa_strdup(shortname); | |
335 | dbn->dbn_dirty = B_FALSE; | |
336 | mutex_init(&dbn->dbn_lock, NULL, MUTEX_DEFAULT, NULL); | |
337 | return (dbn); | |
338 | } | |
339 | ||
340 | /* | |
341 | * Set the fields in the zfs_bookmark_phys_t based on the specified snapshot. | |
342 | */ | |
da536844 | 343 | static void |
30af21b0 PD |
344 | dsl_bookmark_set_phys(zfs_bookmark_phys_t *zbm, dsl_dataset_t *snap) |
345 | { | |
346 | spa_t *spa = dsl_dataset_get_spa(snap); | |
347 | objset_t *mos = spa_get_dsl(spa)->dp_meta_objset; | |
348 | dsl_dataset_phys_t *dsp = dsl_dataset_phys(snap); | |
ff7e405f BB |
349 | |
350 | memset(zbm, 0, sizeof (zfs_bookmark_phys_t)); | |
30af21b0 PD |
351 | zbm->zbm_guid = dsp->ds_guid; |
352 | zbm->zbm_creation_txg = dsp->ds_creation_txg; | |
353 | zbm->zbm_creation_time = dsp->ds_creation_time; | |
354 | zbm->zbm_redaction_obj = 0; | |
355 | ||
356 | /* | |
357 | * If the dataset is encrypted create a larger bookmark to | |
358 | * accommodate the IVset guid. The IVset guid was added | |
359 | * after the encryption feature to prevent a problem with | |
360 | * raw sends. If we encounter an encrypted dataset without | |
361 | * an IVset guid we fall back to a normal bookmark. | |
362 | */ | |
363 | if (snap->ds_dir->dd_crypto_obj != 0 && | |
364 | spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_V2)) { | |
365 | (void) zap_lookup(mos, snap->ds_object, | |
366 | DS_FIELD_IVSET_GUID, sizeof (uint64_t), 1, | |
367 | &zbm->zbm_ivset_guid); | |
368 | } | |
369 | ||
370 | if (spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_WRITTEN)) { | |
371 | zbm->zbm_flags = ZBM_FLAG_SNAPSHOT_EXISTS | ZBM_FLAG_HAS_FBN; | |
372 | zbm->zbm_referenced_bytes_refd = dsp->ds_referenced_bytes; | |
373 | zbm->zbm_compressed_bytes_refd = dsp->ds_compressed_bytes; | |
374 | zbm->zbm_uncompressed_bytes_refd = dsp->ds_uncompressed_bytes; | |
375 | ||
376 | dsl_dataset_t *nextds; | |
377 | VERIFY0(dsl_dataset_hold_obj(snap->ds_dir->dd_pool, | |
378 | dsp->ds_next_snap_obj, FTAG, &nextds)); | |
379 | dsl_deadlist_space(&nextds->ds_deadlist, | |
380 | &zbm->zbm_referenced_freed_before_next_snap, | |
381 | &zbm->zbm_compressed_freed_before_next_snap, | |
382 | &zbm->zbm_uncompressed_freed_before_next_snap); | |
383 | dsl_dataset_rele(nextds, FTAG); | |
30af21b0 PD |
384 | } |
385 | } | |
386 | ||
a73f361f CS |
387 | /* |
388 | * Add dsl_bookmark_node_t `dbn` to the given dataset and increment appropriate | |
389 | * SPA feature counters. | |
390 | */ | |
30af21b0 PD |
391 | void |
392 | dsl_bookmark_node_add(dsl_dataset_t *hds, dsl_bookmark_node_t *dbn, | |
393 | dmu_tx_t *tx) | |
da536844 | 394 | { |
da536844 MA |
395 | dsl_pool_t *dp = dmu_tx_pool(tx); |
396 | objset_t *mos = dp->dp_meta_objset; | |
da536844 | 397 | |
30af21b0 PD |
398 | if (hds->ds_bookmarks_obj == 0) { |
399 | hds->ds_bookmarks_obj = zap_create_norm(mos, | |
400 | U8_TEXTPREP_TOUPPER, DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, | |
401 | tx); | |
402 | spa_feature_incr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx); | |
403 | ||
404 | dsl_dataset_zapify(hds, tx); | |
405 | VERIFY0(zap_add(mos, hds->ds_object, | |
406 | DS_FIELD_BOOKMARK_NAMES, | |
407 | sizeof (hds->ds_bookmarks_obj), 1, | |
408 | &hds->ds_bookmarks_obj, tx)); | |
409 | } | |
da536844 | 410 | |
30af21b0 | 411 | avl_add(&hds->ds_bookmarks, dbn); |
da536844 | 412 | |
30af21b0 PD |
413 | /* |
414 | * To maintain backwards compatibility with software that doesn't | |
415 | * understand SPA_FEATURE_BOOKMARK_V2, we need to use the smallest | |
416 | * possible bookmark size. | |
417 | */ | |
418 | uint64_t bookmark_phys_size = BOOKMARK_PHYS_SIZE_V1; | |
419 | if (spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARK_V2) && | |
420 | (dbn->dbn_phys.zbm_ivset_guid != 0 || dbn->dbn_phys.zbm_flags & | |
421 | ZBM_FLAG_HAS_FBN || dbn->dbn_phys.zbm_redaction_obj != 0)) { | |
422 | bookmark_phys_size = BOOKMARK_PHYS_SIZE_V2; | |
423 | spa_feature_incr(dp->dp_spa, SPA_FEATURE_BOOKMARK_V2, tx); | |
424 | } | |
da536844 | 425 | |
861166b0 AZ |
426 | zfs_bookmark_phys_t zero_phys = { 0 }; |
427 | ASSERT0(memcmp(((char *)&dbn->dbn_phys) + bookmark_phys_size, | |
30af21b0 | 428 | &zero_phys, sizeof (zfs_bookmark_phys_t) - bookmark_phys_size)); |
da536844 | 429 | |
30af21b0 PD |
430 | VERIFY0(zap_add(mos, hds->ds_bookmarks_obj, dbn->dbn_name, |
431 | sizeof (uint64_t), bookmark_phys_size / sizeof (uint64_t), | |
432 | &dbn->dbn_phys, tx)); | |
433 | } | |
434 | ||
435 | /* | |
436 | * If redaction_list is non-null, we create a redacted bookmark and redaction | |
437 | * list, and store the object number of the redaction list in redact_obj. | |
438 | */ | |
439 | static void | |
a73f361f | 440 | dsl_bookmark_create_sync_impl_snap(const char *bookmark, const char *snapshot, |
dd66857d AZ |
441 | dmu_tx_t *tx, uint64_t num_redact_snaps, uint64_t *redact_snaps, |
442 | const void *tag, redaction_list_t **redaction_list) | |
30af21b0 PD |
443 | { |
444 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
445 | objset_t *mos = dp->dp_meta_objset; | |
446 | dsl_dataset_t *snapds, *bmark_fs; | |
447 | char *shortname; | |
448 | boolean_t bookmark_redacted; | |
449 | uint64_t *dsredactsnaps; | |
450 | uint64_t dsnumsnaps; | |
451 | ||
452 | VERIFY0(dsl_dataset_hold(dp, snapshot, FTAG, &snapds)); | |
453 | VERIFY0(dsl_bookmark_hold_ds(dp, bookmark, &bmark_fs, FTAG, | |
454 | &shortname)); | |
455 | ||
456 | dsl_bookmark_node_t *dbn = dsl_bookmark_node_alloc(shortname); | |
457 | dsl_bookmark_set_phys(&dbn->dbn_phys, snapds); | |
458 | ||
459 | bookmark_redacted = dsl_dataset_get_uint64_array_feature(snapds, | |
460 | SPA_FEATURE_REDACTED_DATASETS, &dsnumsnaps, &dsredactsnaps); | |
461 | if (redaction_list != NULL || bookmark_redacted) { | |
462 | redaction_list_t *local_rl; | |
bee9cfb8 | 463 | boolean_t spill = B_FALSE; |
30af21b0 PD |
464 | if (bookmark_redacted) { |
465 | redact_snaps = dsredactsnaps; | |
466 | num_redact_snaps = dsnumsnaps; | |
467 | } | |
bee9cfb8 PD |
468 | int bonuslen = sizeof (redaction_list_phys_t) + |
469 | num_redact_snaps * sizeof (uint64_t); | |
470 | if (bonuslen > dmu_bonus_max()) | |
471 | spill = B_TRUE; | |
30af21b0 PD |
472 | dbn->dbn_phys.zbm_redaction_obj = dmu_object_alloc(mos, |
473 | DMU_OTN_UINT64_METADATA, SPA_OLD_MAXBLOCKSIZE, | |
bee9cfb8 | 474 | DMU_OTN_UINT64_METADATA, spill ? 0 : bonuslen, tx); |
30af21b0 PD |
475 | spa_feature_incr(dp->dp_spa, |
476 | SPA_FEATURE_REDACTION_BOOKMARKS, tx); | |
bee9cfb8 PD |
477 | if (spill) { |
478 | spa_feature_incr(dp->dp_spa, | |
479 | SPA_FEATURE_REDACTION_LIST_SPILL, tx); | |
480 | } | |
30af21b0 PD |
481 | |
482 | VERIFY0(dsl_redaction_list_hold_obj(dp, | |
483 | dbn->dbn_phys.zbm_redaction_obj, tag, &local_rl)); | |
484 | dsl_redaction_list_long_hold(dp, local_rl, tag); | |
485 | ||
bee9cfb8 PD |
486 | if (!spill) { |
487 | ASSERT3U(local_rl->rl_bonus->db_size, >=, bonuslen); | |
488 | dmu_buf_will_dirty(local_rl->rl_bonus, tx); | |
489 | } else { | |
490 | dmu_buf_t *db; | |
491 | VERIFY0(dmu_spill_hold_by_bonus(local_rl->rl_bonus, | |
492 | DB_RF_MUST_SUCCEED, FTAG, &db)); | |
9b1677fb | 493 | dmu_buf_will_fill(db, tx, B_FALSE); |
bee9cfb8 PD |
494 | VERIFY0(dbuf_spill_set_blksz(db, P2ROUNDUP(bonuslen, |
495 | SPA_MINBLOCKSIZE), tx)); | |
496 | local_rl->rl_phys = db->db_data; | |
497 | local_rl->rl_dbuf = db; | |
498 | } | |
861166b0 | 499 | memcpy(local_rl->rl_phys->rlp_snaps, redact_snaps, |
30af21b0 PD |
500 | sizeof (uint64_t) * num_redact_snaps); |
501 | local_rl->rl_phys->rlp_num_snaps = num_redact_snaps; | |
502 | if (bookmark_redacted) { | |
503 | ASSERT3P(redaction_list, ==, NULL); | |
504 | local_rl->rl_phys->rlp_last_blkid = UINT64_MAX; | |
505 | local_rl->rl_phys->rlp_last_object = UINT64_MAX; | |
506 | dsl_redaction_list_long_rele(local_rl, tag); | |
507 | dsl_redaction_list_rele(local_rl, tag); | |
508 | } else { | |
509 | *redaction_list = local_rl; | |
f00ab3f2 | 510 | } |
30af21b0 PD |
511 | } |
512 | ||
513 | if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN) { | |
514 | spa_feature_incr(dp->dp_spa, | |
515 | SPA_FEATURE_BOOKMARK_WRITTEN, tx); | |
516 | } | |
f00ab3f2 | 517 | |
30af21b0 | 518 | dsl_bookmark_node_add(bmark_fs, dbn, tx); |
da536844 | 519 | |
30af21b0 PD |
520 | spa_history_log_internal_ds(bmark_fs, "bookmark", tx, |
521 | "name=%s creation_txg=%llu target_snap=%llu redact_obj=%llu", | |
522 | shortname, (longlong_t)dbn->dbn_phys.zbm_creation_txg, | |
523 | (longlong_t)snapds->ds_object, | |
524 | (longlong_t)dbn->dbn_phys.zbm_redaction_obj); | |
da536844 | 525 | |
30af21b0 PD |
526 | dsl_dataset_rele(bmark_fs, FTAG); |
527 | dsl_dataset_rele(snapds, FTAG); | |
528 | } | |
529 | ||
a73f361f | 530 | |
30af21b0 | 531 | static void |
a73f361f CS |
532 | dsl_bookmark_create_sync_impl_book( |
533 | const char *new_name, const char *source_name, dmu_tx_t *tx) | |
534 | { | |
535 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
536 | dsl_dataset_t *bmark_fs_source, *bmark_fs_new; | |
537 | char *source_shortname, *new_shortname; | |
538 | zfs_bookmark_phys_t source_phys; | |
539 | ||
540 | VERIFY0(dsl_bookmark_hold_ds(dp, source_name, &bmark_fs_source, FTAG, | |
541 | &source_shortname)); | |
542 | VERIFY0(dsl_bookmark_hold_ds(dp, new_name, &bmark_fs_new, FTAG, | |
543 | &new_shortname)); | |
544 | ||
545 | /* | |
546 | * create a copy of the source bookmark by copying most of its members | |
547 | * | |
548 | * Caveat: bookmarking a redaction bookmark yields a normal bookmark | |
549 | * ----------------------------------------------------------------- | |
550 | * Reasoning: | |
551 | * - The zbm_redaction_obj would be referred to by both source and new | |
552 | * bookmark, but would be destroyed once either source or new is | |
bf169e9f | 553 | * destroyed, resulting in use-after-free of the referred object. |
a73f361f CS |
554 | * - User expectation when issuing the `zfs bookmark` command is that |
555 | * a normal bookmark of the source is created | |
556 | * | |
557 | * Design Alternatives For Full Redaction Bookmark Copying: | |
558 | * - reference-count the redaction object => would require on-disk | |
559 | * format change for existing redaction objects | |
560 | * - Copy the redaction object => cannot be done in syncing context | |
561 | * because the redaction object might be too large | |
562 | */ | |
563 | ||
564 | VERIFY0(dsl_bookmark_lookup_impl(bmark_fs_source, source_shortname, | |
565 | &source_phys)); | |
566 | dsl_bookmark_node_t *new_dbn = dsl_bookmark_node_alloc(new_shortname); | |
567 | ||
568 | memcpy(&new_dbn->dbn_phys, &source_phys, sizeof (source_phys)); | |
569 | new_dbn->dbn_phys.zbm_redaction_obj = 0; | |
570 | ||
571 | /* update feature counters */ | |
572 | if (new_dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN) { | |
573 | spa_feature_incr(dp->dp_spa, | |
574 | SPA_FEATURE_BOOKMARK_WRITTEN, tx); | |
575 | } | |
576 | /* no need for redaction bookmark counter; nulled zbm_redaction_obj */ | |
577 | /* dsl_bookmark_node_add bumps bookmarks and v2-bookmarks counter */ | |
578 | ||
579 | /* | |
580 | * write new bookmark | |
581 | * | |
582 | * Note that dsl_bookmark_lookup_impl guarantees that, if source is a | |
583 | * v1 bookmark, the v2-only fields are zeroed. | |
584 | * And dsl_bookmark_node_add writes back a v1-sized bookmark if | |
585 | * v2 bookmarks are disabled and/or v2-only fields are zeroed. | |
586 | * => bookmark copying works on pre-bookmark-v2 pools | |
587 | */ | |
588 | dsl_bookmark_node_add(bmark_fs_new, new_dbn, tx); | |
589 | ||
590 | spa_history_log_internal_ds(bmark_fs_source, "bookmark", tx, | |
591 | "name=%s creation_txg=%llu source_guid=%llu", | |
592 | new_shortname, (longlong_t)new_dbn->dbn_phys.zbm_creation_txg, | |
593 | (longlong_t)source_phys.zbm_guid); | |
594 | ||
595 | dsl_dataset_rele(bmark_fs_source, FTAG); | |
596 | dsl_dataset_rele(bmark_fs_new, FTAG); | |
597 | } | |
598 | ||
599 | void | |
30af21b0 PD |
600 | dsl_bookmark_create_sync(void *arg, dmu_tx_t *tx) |
601 | { | |
602 | dsl_bookmark_create_arg_t *dbca = arg; | |
603 | ||
604 | ASSERT(spa_feature_is_enabled(dmu_tx_pool(tx)->dp_spa, | |
605 | SPA_FEATURE_BOOKMARKS)); | |
606 | ||
607 | for (nvpair_t *pair = nvlist_next_nvpair(dbca->dbca_bmarks, NULL); | |
608 | pair != NULL; pair = nvlist_next_nvpair(dbca->dbca_bmarks, pair)) { | |
a73f361f | 609 | |
d1807f16 RY |
610 | const char *new = nvpair_name(pair); |
611 | const char *source = fnvpair_value_string(pair); | |
a73f361f CS |
612 | |
613 | if (strchr(source, '@') != NULL) { | |
614 | dsl_bookmark_create_sync_impl_snap(new, source, tx, | |
615 | 0, NULL, NULL, NULL); | |
616 | } else if (strchr(source, '#') != NULL) { | |
617 | dsl_bookmark_create_sync_impl_book(new, source, tx); | |
618 | } else { | |
619 | panic("unreachable code"); | |
620 | } | |
621 | ||
da536844 MA |
622 | } |
623 | } | |
624 | ||
625 | /* | |
626 | * The bookmarks must all be in the same pool. | |
627 | */ | |
628 | int | |
629 | dsl_bookmark_create(nvlist_t *bmarks, nvlist_t *errors) | |
630 | { | |
631 | nvpair_t *pair; | |
632 | dsl_bookmark_create_arg_t dbca; | |
633 | ||
634 | pair = nvlist_next_nvpair(bmarks, NULL); | |
635 | if (pair == NULL) | |
636 | return (0); | |
637 | ||
638 | dbca.dbca_bmarks = bmarks; | |
639 | dbca.dbca_errors = errors; | |
640 | ||
641 | return (dsl_sync_task(nvpair_name(pair), dsl_bookmark_create_check, | |
3d45fdd6 MA |
642 | dsl_bookmark_create_sync, &dbca, |
643 | fnvlist_num_pairs(bmarks), ZFS_SPACE_CHECK_NORMAL)); | |
da536844 MA |
644 | } |
645 | ||
30af21b0 PD |
646 | static int |
647 | dsl_bookmark_create_redacted_check(void *arg, dmu_tx_t *tx) | |
da536844 | 648 | { |
30af21b0 PD |
649 | dsl_bookmark_create_redacted_arg_t *dbcra = arg; |
650 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
30af21b0 | 651 | int rv = 0; |
da536844 | 652 | |
30af21b0 PD |
653 | if (!spa_feature_is_enabled(dp->dp_spa, |
654 | SPA_FEATURE_REDACTION_BOOKMARKS)) | |
655 | return (SET_ERROR(ENOTSUP)); | |
656 | /* | |
bee9cfb8 PD |
657 | * If the list of redact snaps will not fit in the bonus buffer (or |
658 | * spill block, with the REDACTION_LIST_SPILL feature) with the | |
659 | * furthest reached object and offset, fail. | |
30af21b0 | 660 | */ |
bee9cfb8 PD |
661 | uint64_t snaplimit = ((spa_feature_is_enabled(dp->dp_spa, |
662 | SPA_FEATURE_REDACTION_LIST_SPILL) ? spa_maxblocksize(dp->dp_spa) : | |
663 | dmu_bonus_max()) - | |
664 | sizeof (redaction_list_phys_t)) / sizeof (uint64_t); | |
665 | if (dbcra->dbcra_numsnaps > snaplimit) | |
30af21b0 PD |
666 | return (SET_ERROR(E2BIG)); |
667 | ||
a73f361f CS |
668 | if (dsl_bookmark_create_nvl_validate_pair( |
669 | dbcra->dbcra_bmark, dbcra->dbcra_snap) != 0) | |
670 | return (SET_ERROR(EINVAL)); | |
671 | ||
672 | rv = dsl_bookmark_create_check_impl(dp, | |
673 | dbcra->dbcra_bmark, dbcra->dbcra_snap); | |
30af21b0 PD |
674 | return (rv); |
675 | } | |
da536844 | 676 | |
30af21b0 PD |
677 | static void |
678 | dsl_bookmark_create_redacted_sync(void *arg, dmu_tx_t *tx) | |
679 | { | |
680 | dsl_bookmark_create_redacted_arg_t *dbcra = arg; | |
a73f361f CS |
681 | dsl_bookmark_create_sync_impl_snap(dbcra->dbcra_bmark, |
682 | dbcra->dbcra_snap, tx, dbcra->dbcra_numsnaps, dbcra->dbcra_snaps, | |
683 | dbcra->dbcra_tag, dbcra->dbcra_rl); | |
30af21b0 | 684 | } |
da536844 | 685 | |
30af21b0 PD |
686 | int |
687 | dsl_bookmark_create_redacted(const char *bookmark, const char *snapshot, | |
dd66857d AZ |
688 | uint64_t numsnaps, uint64_t *snapguids, const void *tag, |
689 | redaction_list_t **rl) | |
30af21b0 PD |
690 | { |
691 | dsl_bookmark_create_redacted_arg_t dbcra; | |
692 | ||
693 | dbcra.dbcra_bmark = bookmark; | |
694 | dbcra.dbcra_snap = snapshot; | |
695 | dbcra.dbcra_rl = rl; | |
696 | dbcra.dbcra_numsnaps = numsnaps; | |
697 | dbcra.dbcra_snaps = snapguids; | |
698 | dbcra.dbcra_tag = tag; | |
699 | ||
700 | return (dsl_sync_task(bookmark, dsl_bookmark_create_redacted_check, | |
701 | dsl_bookmark_create_redacted_sync, &dbcra, 5, | |
702 | ZFS_SPACE_CHECK_NORMAL)); | |
703 | } | |
da536844 | 704 | |
30af21b0 PD |
705 | /* |
706 | * Retrieve the list of properties given in the 'props' nvlist for a bookmark. | |
707 | * If 'props' is NULL, retrieves all properties. | |
708 | */ | |
709 | static void | |
710 | dsl_bookmark_fetch_props(dsl_pool_t *dp, zfs_bookmark_phys_t *bmark_phys, | |
711 | nvlist_t *props, nvlist_t *out_props) | |
712 | { | |
713 | ASSERT3P(dp, !=, NULL); | |
714 | ASSERT3P(bmark_phys, !=, NULL); | |
715 | ASSERT3P(out_props, !=, NULL); | |
716 | ASSERT(RRW_LOCK_HELD(&dp->dp_config_rwlock)); | |
717 | ||
718 | if (props == NULL || nvlist_exists(props, | |
719 | zfs_prop_to_name(ZFS_PROP_GUID))) { | |
720 | dsl_prop_nvlist_add_uint64(out_props, | |
721 | ZFS_PROP_GUID, bmark_phys->zbm_guid); | |
722 | } | |
723 | if (props == NULL || nvlist_exists(props, | |
724 | zfs_prop_to_name(ZFS_PROP_CREATETXG))) { | |
725 | dsl_prop_nvlist_add_uint64(out_props, | |
726 | ZFS_PROP_CREATETXG, bmark_phys->zbm_creation_txg); | |
727 | } | |
728 | if (props == NULL || nvlist_exists(props, | |
729 | zfs_prop_to_name(ZFS_PROP_CREATION))) { | |
730 | dsl_prop_nvlist_add_uint64(out_props, | |
731 | ZFS_PROP_CREATION, bmark_phys->zbm_creation_time); | |
732 | } | |
733 | if (props == NULL || nvlist_exists(props, | |
734 | zfs_prop_to_name(ZFS_PROP_IVSET_GUID))) { | |
735 | dsl_prop_nvlist_add_uint64(out_props, | |
736 | ZFS_PROP_IVSET_GUID, bmark_phys->zbm_ivset_guid); | |
737 | } | |
738 | if (bmark_phys->zbm_flags & ZBM_FLAG_HAS_FBN) { | |
739 | if (props == NULL || nvlist_exists(props, | |
740 | zfs_prop_to_name(ZFS_PROP_REFERENCED))) { | |
da536844 | 741 | dsl_prop_nvlist_add_uint64(out_props, |
30af21b0 PD |
742 | ZFS_PROP_REFERENCED, |
743 | bmark_phys->zbm_referenced_bytes_refd); | |
da536844 | 744 | } |
30af21b0 PD |
745 | if (props == NULL || nvlist_exists(props, |
746 | zfs_prop_to_name(ZFS_PROP_LOGICALREFERENCED))) { | |
da536844 | 747 | dsl_prop_nvlist_add_uint64(out_props, |
30af21b0 PD |
748 | ZFS_PROP_LOGICALREFERENCED, |
749 | bmark_phys->zbm_uncompressed_bytes_refd); | |
da536844 | 750 | } |
30af21b0 PD |
751 | if (props == NULL || nvlist_exists(props, |
752 | zfs_prop_to_name(ZFS_PROP_REFRATIO))) { | |
753 | uint64_t ratio = | |
754 | bmark_phys->zbm_compressed_bytes_refd == 0 ? 100 : | |
755 | bmark_phys->zbm_uncompressed_bytes_refd * 100 / | |
756 | bmark_phys->zbm_compressed_bytes_refd; | |
da536844 | 757 | dsl_prop_nvlist_add_uint64(out_props, |
30af21b0 | 758 | ZFS_PROP_REFRATIO, ratio); |
da536844 | 759 | } |
30af21b0 PD |
760 | } |
761 | ||
762 | if ((props == NULL || nvlist_exists(props, "redact_snaps") || | |
763 | nvlist_exists(props, "redact_complete")) && | |
764 | bmark_phys->zbm_redaction_obj != 0) { | |
765 | redaction_list_t *rl; | |
766 | int err = dsl_redaction_list_hold_obj(dp, | |
767 | bmark_phys->zbm_redaction_obj, FTAG, &rl); | |
768 | if (err == 0) { | |
769 | if (nvlist_exists(props, "redact_snaps")) { | |
770 | nvlist_t *nvl; | |
771 | nvl = fnvlist_alloc(); | |
772 | fnvlist_add_uint64_array(nvl, ZPROP_VALUE, | |
773 | rl->rl_phys->rlp_snaps, | |
774 | rl->rl_phys->rlp_num_snaps); | |
775 | fnvlist_add_nvlist(out_props, "redact_snaps", | |
776 | nvl); | |
777 | nvlist_free(nvl); | |
778 | } | |
779 | if (nvlist_exists(props, "redact_complete")) { | |
780 | nvlist_t *nvl; | |
781 | nvl = fnvlist_alloc(); | |
782 | fnvlist_add_boolean_value(nvl, ZPROP_VALUE, | |
783 | rl->rl_phys->rlp_last_blkid == UINT64_MAX && | |
784 | rl->rl_phys->rlp_last_object == UINT64_MAX); | |
785 | fnvlist_add_nvlist(out_props, "redact_complete", | |
786 | nvl); | |
787 | nvlist_free(nvl); | |
788 | } | |
789 | dsl_redaction_list_rele(rl, FTAG); | |
f00ab3f2 | 790 | } |
30af21b0 PD |
791 | } |
792 | } | |
793 | ||
794 | int | |
795 | dsl_get_bookmarks_impl(dsl_dataset_t *ds, nvlist_t *props, nvlist_t *outnvl) | |
796 | { | |
797 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
da536844 | 798 | |
30af21b0 PD |
799 | ASSERT(dsl_pool_config_held(dp)); |
800 | ||
801 | if (dsl_dataset_is_snapshot(ds)) | |
802 | return (SET_ERROR(EINVAL)); | |
803 | ||
804 | for (dsl_bookmark_node_t *dbn = avl_first(&ds->ds_bookmarks); | |
805 | dbn != NULL; dbn = AVL_NEXT(&ds->ds_bookmarks, dbn)) { | |
806 | nvlist_t *out_props = fnvlist_alloc(); | |
807 | ||
808 | dsl_bookmark_fetch_props(dp, &dbn->dbn_phys, props, out_props); | |
809 | ||
810 | fnvlist_add_nvlist(outnvl, dbn->dbn_name, out_props); | |
da536844 MA |
811 | fnvlist_free(out_props); |
812 | } | |
30af21b0 PD |
813 | return (0); |
814 | } | |
815 | ||
816 | /* | |
817 | * Comparison func for ds_bookmarks AVL tree. We sort the bookmarks by | |
818 | * their TXG, then by their FBN-ness. The "FBN-ness" component ensures | |
819 | * that all bookmarks at the same TXG that HAS_FBN are adjacent, which | |
820 | * dsl_bookmark_destroy_sync_impl() depends on. Note that there may be | |
821 | * multiple bookmarks at the same TXG (with the same FBN-ness). In this | |
822 | * case we differentiate them by an arbitrary metric (in this case, | |
823 | * their names). | |
824 | */ | |
825 | static int | |
826 | dsl_bookmark_compare(const void *l, const void *r) | |
827 | { | |
828 | const dsl_bookmark_node_t *ldbn = l; | |
829 | const dsl_bookmark_node_t *rdbn = r; | |
830 | ||
ca577779 | 831 | int64_t cmp = TREE_CMP(ldbn->dbn_phys.zbm_creation_txg, |
30af21b0 PD |
832 | rdbn->dbn_phys.zbm_creation_txg); |
833 | if (likely(cmp)) | |
834 | return (cmp); | |
ca577779 | 835 | cmp = TREE_CMP((ldbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN), |
30af21b0 PD |
836 | (rdbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)); |
837 | if (likely(cmp)) | |
838 | return (cmp); | |
839 | cmp = strcmp(ldbn->dbn_name, rdbn->dbn_name); | |
ca577779 | 840 | return (TREE_ISIGN(cmp)); |
30af21b0 PD |
841 | } |
842 | ||
843 | /* | |
844 | * Cache this (head) dataset's bookmarks in the ds_bookmarks AVL tree. | |
845 | */ | |
846 | int | |
847 | dsl_bookmark_init_ds(dsl_dataset_t *ds) | |
848 | { | |
849 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
850 | objset_t *mos = dp->dp_meta_objset; | |
851 | ||
852 | ASSERT(!ds->ds_is_snapshot); | |
853 | ||
854 | avl_create(&ds->ds_bookmarks, dsl_bookmark_compare, | |
855 | sizeof (dsl_bookmark_node_t), | |
856 | offsetof(dsl_bookmark_node_t, dbn_node)); | |
857 | ||
858 | if (!dsl_dataset_is_zapified(ds)) | |
859 | return (0); | |
860 | ||
861 | int zaperr = zap_lookup(mos, ds->ds_object, DS_FIELD_BOOKMARK_NAMES, | |
862 | sizeof (ds->ds_bookmarks_obj), 1, &ds->ds_bookmarks_obj); | |
863 | if (zaperr == ENOENT) | |
864 | return (0); | |
865 | if (zaperr != 0) | |
866 | return (zaperr); | |
867 | ||
868 | if (ds->ds_bookmarks_obj == 0) | |
869 | return (0); | |
870 | ||
871 | int err = 0; | |
872 | zap_cursor_t zc; | |
873 | zap_attribute_t attr; | |
874 | ||
875 | for (zap_cursor_init(&zc, mos, ds->ds_bookmarks_obj); | |
876 | (err = zap_cursor_retrieve(&zc, &attr)) == 0; | |
877 | zap_cursor_advance(&zc)) { | |
878 | dsl_bookmark_node_t *dbn = | |
879 | dsl_bookmark_node_alloc(attr.za_name); | |
880 | ||
881 | err = dsl_bookmark_lookup_impl(ds, | |
882 | dbn->dbn_name, &dbn->dbn_phys); | |
883 | ASSERT3U(err, !=, ENOENT); | |
884 | if (err != 0) { | |
885 | kmem_free(dbn, sizeof (*dbn)); | |
886 | break; | |
887 | } | |
888 | avl_add(&ds->ds_bookmarks, dbn); | |
889 | } | |
da536844 | 890 | zap_cursor_fini(&zc); |
30af21b0 PD |
891 | if (err == ENOENT) |
892 | err = 0; | |
da536844 MA |
893 | return (err); |
894 | } | |
895 | ||
30af21b0 PD |
896 | void |
897 | dsl_bookmark_fini_ds(dsl_dataset_t *ds) | |
898 | { | |
899 | void *cookie = NULL; | |
900 | dsl_bookmark_node_t *dbn; | |
901 | ||
902 | if (ds->ds_is_snapshot) | |
903 | return; | |
904 | ||
905 | while ((dbn = avl_destroy_nodes(&ds->ds_bookmarks, &cookie)) != NULL) { | |
906 | spa_strfree(dbn->dbn_name); | |
907 | mutex_destroy(&dbn->dbn_lock); | |
908 | kmem_free(dbn, sizeof (*dbn)); | |
909 | } | |
910 | avl_destroy(&ds->ds_bookmarks); | |
911 | } | |
912 | ||
da536844 MA |
913 | /* |
914 | * Retrieve the bookmarks that exist in the specified dataset, and the | |
915 | * requested properties of each bookmark. | |
916 | * | |
917 | * The "props" nvlist specifies which properties are requested. | |
918 | * See lzc_get_bookmarks() for the list of valid properties. | |
919 | */ | |
920 | int | |
921 | dsl_get_bookmarks(const char *dsname, nvlist_t *props, nvlist_t *outnvl) | |
922 | { | |
923 | dsl_pool_t *dp; | |
924 | dsl_dataset_t *ds; | |
925 | int err; | |
926 | ||
927 | err = dsl_pool_hold(dsname, FTAG, &dp); | |
928 | if (err != 0) | |
929 | return (err); | |
930 | err = dsl_dataset_hold(dp, dsname, FTAG, &ds); | |
931 | if (err != 0) { | |
932 | dsl_pool_rele(dp, FTAG); | |
933 | return (err); | |
934 | } | |
935 | ||
936 | err = dsl_get_bookmarks_impl(ds, props, outnvl); | |
937 | ||
938 | dsl_dataset_rele(ds, FTAG); | |
939 | dsl_pool_rele(dp, FTAG); | |
940 | return (err); | |
941 | } | |
942 | ||
30af21b0 PD |
943 | /* |
944 | * Retrieve all properties for a single bookmark in the given dataset. | |
945 | */ | |
946 | int | |
947 | dsl_get_bookmark_props(const char *dsname, const char *bmname, nvlist_t *props) | |
948 | { | |
949 | dsl_pool_t *dp; | |
950 | dsl_dataset_t *ds; | |
951 | zfs_bookmark_phys_t bmark_phys = { 0 }; | |
952 | int err; | |
953 | ||
954 | err = dsl_pool_hold(dsname, FTAG, &dp); | |
955 | if (err != 0) | |
956 | return (err); | |
957 | err = dsl_dataset_hold(dp, dsname, FTAG, &ds); | |
958 | if (err != 0) { | |
959 | dsl_pool_rele(dp, FTAG); | |
960 | return (err); | |
961 | } | |
962 | ||
963 | err = dsl_bookmark_lookup_impl(ds, bmname, &bmark_phys); | |
964 | if (err != 0) | |
965 | goto out; | |
966 | ||
967 | dsl_bookmark_fetch_props(dp, &bmark_phys, NULL, props); | |
968 | out: | |
969 | dsl_dataset_rele(ds, FTAG); | |
970 | dsl_pool_rele(dp, FTAG); | |
971 | return (err); | |
972 | } | |
973 | ||
da536844 MA |
974 | typedef struct dsl_bookmark_destroy_arg { |
975 | nvlist_t *dbda_bmarks; | |
976 | nvlist_t *dbda_success; | |
977 | nvlist_t *dbda_errors; | |
978 | } dsl_bookmark_destroy_arg_t; | |
979 | ||
30af21b0 PD |
980 | static void |
981 | dsl_bookmark_destroy_sync_impl(dsl_dataset_t *ds, const char *name, | |
982 | dmu_tx_t *tx) | |
da536844 MA |
983 | { |
984 | objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
30af21b0 | 985 | uint64_t bmark_zapobj = ds->ds_bookmarks_obj; |
9b7b9cd3 | 986 | matchtype_t mt = 0; |
f00ab3f2 | 987 | uint64_t int_size, num_ints; |
30af21b0 PD |
988 | /* |
989 | * 'search' must be zeroed so that dbn_flags (which is used in | |
990 | * dsl_bookmark_compare()) will be zeroed even if the on-disk | |
991 | * (in ZAP) bookmark is shorter than offsetof(dbn_flags). | |
992 | */ | |
993 | dsl_bookmark_node_t search = { 0 }; | |
994 | char realname[ZFS_MAX_DATASET_NAME_LEN]; | |
995 | ||
996 | /* | |
997 | * Find the real name of this bookmark, which may be different | |
998 | * from the given name if the dataset is case-insensitive. Then | |
999 | * use the real name to find the node in the ds_bookmarks AVL tree. | |
1000 | */ | |
da536844 | 1001 | |
d683ddbb | 1002 | if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) |
9b7b9cd3 | 1003 | mt = MT_NORMALIZE; |
da536844 | 1004 | |
30af21b0 | 1005 | VERIFY0(zap_length(mos, bmark_zapobj, name, &int_size, &num_ints)); |
f00ab3f2 TC |
1006 | |
1007 | ASSERT3U(int_size, ==, sizeof (uint64_t)); | |
1008 | ||
1009 | if (num_ints * int_size > BOOKMARK_PHYS_SIZE_V1) { | |
1010 | spa_feature_decr(dmu_objset_spa(mos), | |
1011 | SPA_FEATURE_BOOKMARK_V2, tx); | |
1012 | } | |
30af21b0 PD |
1013 | VERIFY0(zap_lookup_norm(mos, bmark_zapobj, name, sizeof (uint64_t), |
1014 | num_ints, &search.dbn_phys, mt, realname, sizeof (realname), NULL)); | |
1015 | ||
1016 | search.dbn_name = realname; | |
1017 | dsl_bookmark_node_t *dbn = avl_find(&ds->ds_bookmarks, &search, NULL); | |
1018 | ASSERT(dbn != NULL); | |
1019 | ||
1020 | if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN) { | |
1021 | /* | |
1022 | * If this bookmark HAS_FBN, and it is before the most | |
1023 | * recent snapshot, then its TXG is a key in the head's | |
1024 | * deadlist (and all clones' heads' deadlists). If this is | |
1025 | * the last thing keeping the key (i.e. there are no more | |
1026 | * bookmarks with HAS_FBN at this TXG, and there is no | |
1027 | * snapshot at this TXG), then remove the key. | |
1028 | * | |
1029 | * Note that this algorithm depends on ds_bookmarks being | |
1030 | * sorted such that all bookmarks at the same TXG with | |
1031 | * HAS_FBN are adjacent (with no non-HAS_FBN bookmarks | |
1032 | * at the same TXG in between them). If this were not | |
1033 | * the case, we would need to examine *all* bookmarks | |
1034 | * at this TXG, rather than just the adjacent ones. | |
1035 | */ | |
1036 | ||
1037 | dsl_bookmark_node_t *dbn_prev = | |
1038 | AVL_PREV(&ds->ds_bookmarks, dbn); | |
1039 | dsl_bookmark_node_t *dbn_next = | |
1040 | AVL_NEXT(&ds->ds_bookmarks, dbn); | |
1041 | ||
1042 | boolean_t more_bookmarks_at_this_txg = | |
1043 | (dbn_prev != NULL && dbn_prev->dbn_phys.zbm_creation_txg == | |
1044 | dbn->dbn_phys.zbm_creation_txg && | |
1045 | (dbn_prev->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)) || | |
1046 | (dbn_next != NULL && dbn_next->dbn_phys.zbm_creation_txg == | |
1047 | dbn->dbn_phys.zbm_creation_txg && | |
1048 | (dbn_next->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)); | |
1049 | ||
1050 | if (!(dbn->dbn_phys.zbm_flags & ZBM_FLAG_SNAPSHOT_EXISTS) && | |
1051 | !more_bookmarks_at_this_txg && | |
1052 | dbn->dbn_phys.zbm_creation_txg < | |
1053 | dsl_dataset_phys(ds)->ds_prev_snap_txg) { | |
1054 | dsl_dir_remove_clones_key(ds->ds_dir, | |
1055 | dbn->dbn_phys.zbm_creation_txg, tx); | |
1056 | dsl_deadlist_remove_key(&ds->ds_deadlist, | |
1057 | dbn->dbn_phys.zbm_creation_txg, tx); | |
1058 | } | |
1059 | ||
1060 | spa_feature_decr(dmu_objset_spa(mos), | |
1061 | SPA_FEATURE_BOOKMARK_WRITTEN, tx); | |
1062 | } | |
1063 | ||
1064 | if (dbn->dbn_phys.zbm_redaction_obj != 0) { | |
bee9cfb8 PD |
1065 | dnode_t *rl; |
1066 | VERIFY0(dnode_hold(mos, | |
1067 | dbn->dbn_phys.zbm_redaction_obj, FTAG, &rl)); | |
1068 | if (rl->dn_have_spill) { | |
1069 | spa_feature_decr(dmu_objset_spa(mos), | |
1070 | SPA_FEATURE_REDACTION_LIST_SPILL, tx); | |
1071 | } | |
1072 | dnode_rele(rl, FTAG); | |
30af21b0 PD |
1073 | VERIFY0(dmu_object_free(mos, |
1074 | dbn->dbn_phys.zbm_redaction_obj, tx)); | |
1075 | spa_feature_decr(dmu_objset_spa(mos), | |
1076 | SPA_FEATURE_REDACTION_BOOKMARKS, tx); | |
1077 | } | |
1078 | ||
1079 | avl_remove(&ds->ds_bookmarks, dbn); | |
1080 | spa_strfree(dbn->dbn_name); | |
1081 | mutex_destroy(&dbn->dbn_lock); | |
1082 | kmem_free(dbn, sizeof (*dbn)); | |
f00ab3f2 | 1083 | |
30af21b0 | 1084 | VERIFY0(zap_remove_norm(mos, bmark_zapobj, name, mt, tx)); |
da536844 MA |
1085 | } |
1086 | ||
1087 | static int | |
1088 | dsl_bookmark_destroy_check(void *arg, dmu_tx_t *tx) | |
1089 | { | |
1090 | dsl_bookmark_destroy_arg_t *dbda = arg; | |
1091 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1092 | int rv = 0; | |
da536844 | 1093 | |
c6f6767e MA |
1094 | ASSERT(nvlist_empty(dbda->dbda_success)); |
1095 | ASSERT(nvlist_empty(dbda->dbda_errors)); | |
1096 | ||
da536844 MA |
1097 | if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARKS)) |
1098 | return (0); | |
1099 | ||
1c27024e | 1100 | for (nvpair_t *pair = nvlist_next_nvpair(dbda->dbda_bmarks, NULL); |
da536844 MA |
1101 | pair != NULL; pair = nvlist_next_nvpair(dbda->dbda_bmarks, pair)) { |
1102 | const char *fullname = nvpair_name(pair); | |
1103 | dsl_dataset_t *ds; | |
1104 | zfs_bookmark_phys_t bm; | |
1105 | int error; | |
1106 | char *shortname; | |
1107 | ||
1108 | error = dsl_bookmark_hold_ds(dp, fullname, &ds, | |
1109 | FTAG, &shortname); | |
1110 | if (error == ENOENT) { | |
1111 | /* ignore it; the bookmark is "already destroyed" */ | |
1112 | continue; | |
1113 | } | |
1114 | if (error == 0) { | |
30af21b0 | 1115 | error = dsl_bookmark_lookup_impl(ds, shortname, &bm); |
da536844 MA |
1116 | dsl_dataset_rele(ds, FTAG); |
1117 | if (error == ESRCH) { | |
1118 | /* | |
1119 | * ignore it; the bookmark is | |
1120 | * "already destroyed" | |
1121 | */ | |
1122 | continue; | |
1123 | } | |
30af21b0 PD |
1124 | if (error == 0 && bm.zbm_redaction_obj != 0) { |
1125 | redaction_list_t *rl = NULL; | |
1126 | error = dsl_redaction_list_hold_obj(tx->tx_pool, | |
1127 | bm.zbm_redaction_obj, FTAG, &rl); | |
1128 | if (error == ENOENT) { | |
1129 | error = 0; | |
1130 | } else if (error == 0 && | |
1131 | dsl_redaction_list_long_held(rl)) { | |
1132 | error = SET_ERROR(EBUSY); | |
1133 | } | |
1134 | if (rl != NULL) { | |
1135 | dsl_redaction_list_rele(rl, FTAG); | |
1136 | } | |
1137 | } | |
da536844 MA |
1138 | } |
1139 | if (error == 0) { | |
c6f6767e MA |
1140 | if (dmu_tx_is_syncing(tx)) { |
1141 | fnvlist_add_boolean(dbda->dbda_success, | |
1142 | fullname); | |
1143 | } | |
da536844 MA |
1144 | } else { |
1145 | fnvlist_add_int32(dbda->dbda_errors, fullname, error); | |
1146 | rv = error; | |
1147 | } | |
1148 | } | |
1149 | return (rv); | |
1150 | } | |
1151 | ||
1152 | static void | |
1153 | dsl_bookmark_destroy_sync(void *arg, dmu_tx_t *tx) | |
1154 | { | |
1155 | dsl_bookmark_destroy_arg_t *dbda = arg; | |
1156 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1157 | objset_t *mos = dp->dp_meta_objset; | |
da536844 | 1158 | |
1c27024e | 1159 | for (nvpair_t *pair = nvlist_next_nvpair(dbda->dbda_success, NULL); |
da536844 MA |
1160 | pair != NULL; pair = nvlist_next_nvpair(dbda->dbda_success, pair)) { |
1161 | dsl_dataset_t *ds; | |
1162 | char *shortname; | |
1163 | uint64_t zap_cnt; | |
1164 | ||
1165 | VERIFY0(dsl_bookmark_hold_ds(dp, nvpair_name(pair), | |
1166 | &ds, FTAG, &shortname)); | |
30af21b0 | 1167 | dsl_bookmark_destroy_sync_impl(ds, shortname, tx); |
da536844 MA |
1168 | |
1169 | /* | |
1170 | * If all of this dataset's bookmarks have been destroyed, | |
1171 | * free the zap object and decrement the feature's use count. | |
1172 | */ | |
30af21b0 | 1173 | VERIFY0(zap_count(mos, ds->ds_bookmarks_obj, &zap_cnt)); |
da536844 MA |
1174 | if (zap_cnt == 0) { |
1175 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
30af21b0 PD |
1176 | VERIFY0(zap_destroy(mos, ds->ds_bookmarks_obj, tx)); |
1177 | ds->ds_bookmarks_obj = 0; | |
da536844 MA |
1178 | spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx); |
1179 | VERIFY0(zap_remove(mos, ds->ds_object, | |
1180 | DS_FIELD_BOOKMARK_NAMES, tx)); | |
1181 | } | |
1182 | ||
1183 | spa_history_log_internal_ds(ds, "remove bookmark", tx, | |
1184 | "name=%s", shortname); | |
1185 | ||
1186 | dsl_dataset_rele(ds, FTAG); | |
1187 | } | |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * The bookmarks must all be in the same pool. | |
1192 | */ | |
1193 | int | |
1194 | dsl_bookmark_destroy(nvlist_t *bmarks, nvlist_t *errors) | |
1195 | { | |
1196 | int rv; | |
1197 | dsl_bookmark_destroy_arg_t dbda; | |
1198 | nvpair_t *pair = nvlist_next_nvpair(bmarks, NULL); | |
1199 | if (pair == NULL) | |
1200 | return (0); | |
1201 | ||
1202 | dbda.dbda_bmarks = bmarks; | |
1203 | dbda.dbda_errors = errors; | |
1204 | dbda.dbda_success = fnvlist_alloc(); | |
1205 | ||
1206 | rv = dsl_sync_task(nvpair_name(pair), dsl_bookmark_destroy_check, | |
3d45fdd6 MA |
1207 | dsl_bookmark_destroy_sync, &dbda, fnvlist_num_pairs(bmarks), |
1208 | ZFS_SPACE_CHECK_RESERVED); | |
da536844 MA |
1209 | fnvlist_free(dbda.dbda_success); |
1210 | return (rv); | |
1211 | } | |
30af21b0 PD |
1212 | |
1213 | /* Return B_TRUE if there are any long holds on this dataset. */ | |
1214 | boolean_t | |
1215 | dsl_redaction_list_long_held(redaction_list_t *rl) | |
1216 | { | |
1217 | return (!zfs_refcount_is_zero(&rl->rl_longholds)); | |
1218 | } | |
1219 | ||
1220 | void | |
dd66857d AZ |
1221 | dsl_redaction_list_long_hold(dsl_pool_t *dp, redaction_list_t *rl, |
1222 | const void *tag) | |
30af21b0 PD |
1223 | { |
1224 | ASSERT(dsl_pool_config_held(dp)); | |
1225 | (void) zfs_refcount_add(&rl->rl_longholds, tag); | |
1226 | } | |
1227 | ||
1228 | void | |
dd66857d | 1229 | dsl_redaction_list_long_rele(redaction_list_t *rl, const void *tag) |
30af21b0 PD |
1230 | { |
1231 | (void) zfs_refcount_remove(&rl->rl_longholds, tag); | |
1232 | } | |
1233 | ||
30af21b0 PD |
1234 | static void |
1235 | redaction_list_evict_sync(void *rlu) | |
1236 | { | |
1237 | redaction_list_t *rl = rlu; | |
1238 | zfs_refcount_destroy(&rl->rl_longholds); | |
1239 | ||
1240 | kmem_free(rl, sizeof (redaction_list_t)); | |
1241 | } | |
1242 | ||
1243 | void | |
dd66857d | 1244 | dsl_redaction_list_rele(redaction_list_t *rl, const void *tag) |
30af21b0 | 1245 | { |
bee9cfb8 PD |
1246 | if (rl->rl_bonus != rl->rl_dbuf) |
1247 | dmu_buf_rele(rl->rl_dbuf, tag); | |
1248 | dmu_buf_rele(rl->rl_bonus, tag); | |
30af21b0 PD |
1249 | } |
1250 | ||
1251 | int | |
dd66857d | 1252 | dsl_redaction_list_hold_obj(dsl_pool_t *dp, uint64_t rlobj, const void *tag, |
30af21b0 PD |
1253 | redaction_list_t **rlp) |
1254 | { | |
1255 | objset_t *mos = dp->dp_meta_objset; | |
bee9cfb8 | 1256 | dmu_buf_t *dbuf, *spill_dbuf; |
30af21b0 PD |
1257 | redaction_list_t *rl; |
1258 | int err; | |
1259 | ||
1260 | ASSERT(dsl_pool_config_held(dp)); | |
1261 | ||
1262 | err = dmu_bonus_hold(mos, rlobj, tag, &dbuf); | |
1263 | if (err != 0) | |
1264 | return (err); | |
1265 | ||
1266 | rl = dmu_buf_get_user(dbuf); | |
1267 | if (rl == NULL) { | |
1268 | redaction_list_t *winner = NULL; | |
1269 | ||
1270 | rl = kmem_zalloc(sizeof (redaction_list_t), KM_SLEEP); | |
bee9cfb8 PD |
1271 | rl->rl_bonus = dbuf; |
1272 | if (dmu_spill_hold_existing(dbuf, tag, &spill_dbuf) == 0) { | |
1273 | rl->rl_dbuf = spill_dbuf; | |
1274 | } else { | |
1275 | rl->rl_dbuf = dbuf; | |
1276 | } | |
30af21b0 | 1277 | rl->rl_object = rlobj; |
bee9cfb8 | 1278 | rl->rl_phys = rl->rl_dbuf->db_data; |
30af21b0 PD |
1279 | rl->rl_mos = dp->dp_meta_objset; |
1280 | zfs_refcount_create(&rl->rl_longholds); | |
1281 | dmu_buf_init_user(&rl->rl_dbu, redaction_list_evict_sync, NULL, | |
bee9cfb8 | 1282 | &rl->rl_bonus); |
30af21b0 PD |
1283 | if ((winner = dmu_buf_set_user_ie(dbuf, &rl->rl_dbu)) != NULL) { |
1284 | kmem_free(rl, sizeof (*rl)); | |
1285 | rl = winner; | |
1286 | } | |
1287 | } | |
1288 | *rlp = rl; | |
1289 | return (0); | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * Snapshot ds is being destroyed. | |
1294 | * | |
1295 | * Adjust the "freed_before_next" of any bookmarks between this snap | |
1296 | * and the previous snapshot, because their "next snapshot" is changing. | |
1297 | * | |
1298 | * If there are any bookmarks with HAS_FBN at this snapshot, remove | |
1299 | * their HAS_SNAP flag (note: there can be at most one snapshot of | |
1300 | * each filesystem at a given txg), and return B_TRUE. In this case | |
1301 | * the caller can not remove the key in the deadlist at this TXG, because | |
1302 | * the HAS_FBN bookmarks require the key be there. | |
1303 | * | |
1304 | * Returns B_FALSE if there are no bookmarks with HAS_FBN at this | |
1305 | * snapshot's TXG. In this case the caller can remove the key in the | |
1306 | * deadlist at this TXG. | |
1307 | */ | |
1308 | boolean_t | |
1309 | dsl_bookmark_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx) | |
1310 | { | |
1311 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
1312 | ||
1313 | dsl_dataset_t *head, *next; | |
1314 | VERIFY0(dsl_dataset_hold_obj(dp, | |
1315 | dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &head)); | |
1316 | VERIFY0(dsl_dataset_hold_obj(dp, | |
1317 | dsl_dataset_phys(ds)->ds_next_snap_obj, FTAG, &next)); | |
1318 | ||
1319 | /* | |
1320 | * Find the first bookmark that HAS_FBN at or after the | |
1321 | * previous snapshot. | |
1322 | */ | |
1323 | dsl_bookmark_node_t search = { 0 }; | |
1324 | avl_index_t idx; | |
1325 | search.dbn_phys.zbm_creation_txg = | |
1326 | dsl_dataset_phys(ds)->ds_prev_snap_txg; | |
1327 | search.dbn_phys.zbm_flags = ZBM_FLAG_HAS_FBN; | |
1328 | /* | |
1329 | * The empty-string name can't be in the AVL, and it compares | |
1330 | * before any entries with this TXG. | |
1331 | */ | |
a926aab9 | 1332 | search.dbn_name = (char *)""; |
30af21b0 PD |
1333 | VERIFY3P(avl_find(&head->ds_bookmarks, &search, &idx), ==, NULL); |
1334 | dsl_bookmark_node_t *dbn = | |
1335 | avl_nearest(&head->ds_bookmarks, idx, AVL_AFTER); | |
1336 | ||
1337 | /* | |
1338 | * Iterate over all bookmarks that are at or after the previous | |
1339 | * snapshot, and before this (being deleted) snapshot. Adjust | |
1340 | * their FBN based on their new next snapshot. | |
1341 | */ | |
1342 | for (; dbn != NULL && dbn->dbn_phys.zbm_creation_txg < | |
1343 | dsl_dataset_phys(ds)->ds_creation_txg; | |
1344 | dbn = AVL_NEXT(&head->ds_bookmarks, dbn)) { | |
1345 | if (!(dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)) | |
1346 | continue; | |
1347 | /* | |
1348 | * Increase our FBN by the amount of space that was live | |
1349 | * (referenced) at the time of this bookmark (i.e. | |
1350 | * birth <= zbm_creation_txg), and killed between this | |
1351 | * (being deleted) snapshot and the next snapshot (i.e. | |
1352 | * on the next snapshot's deadlist). (Space killed before | |
1353 | * this are already on our FBN.) | |
1354 | */ | |
1355 | uint64_t referenced, compressed, uncompressed; | |
1356 | dsl_deadlist_space_range(&next->ds_deadlist, | |
1357 | 0, dbn->dbn_phys.zbm_creation_txg, | |
1358 | &referenced, &compressed, &uncompressed); | |
1359 | dbn->dbn_phys.zbm_referenced_freed_before_next_snap += | |
1360 | referenced; | |
1361 | dbn->dbn_phys.zbm_compressed_freed_before_next_snap += | |
1362 | compressed; | |
1363 | dbn->dbn_phys.zbm_uncompressed_freed_before_next_snap += | |
1364 | uncompressed; | |
1365 | VERIFY0(zap_update(dp->dp_meta_objset, head->ds_bookmarks_obj, | |
1366 | dbn->dbn_name, sizeof (uint64_t), | |
1367 | sizeof (zfs_bookmark_phys_t) / sizeof (uint64_t), | |
1368 | &dbn->dbn_phys, tx)); | |
1369 | } | |
1370 | dsl_dataset_rele(next, FTAG); | |
1371 | ||
1372 | /* | |
1373 | * There may be several bookmarks at this txg (the TXG of the | |
1374 | * snapshot being deleted). We need to clear the SNAPSHOT_EXISTS | |
1375 | * flag on all of them, and return TRUE if there is at least 1 | |
1376 | * bookmark here with HAS_FBN (thus preventing the deadlist | |
1377 | * key from being removed). | |
1378 | */ | |
1379 | boolean_t rv = B_FALSE; | |
1380 | for (; dbn != NULL && dbn->dbn_phys.zbm_creation_txg == | |
1381 | dsl_dataset_phys(ds)->ds_creation_txg; | |
1382 | dbn = AVL_NEXT(&head->ds_bookmarks, dbn)) { | |
1383 | if (!(dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)) { | |
1384 | ASSERT(!(dbn->dbn_phys.zbm_flags & | |
1385 | ZBM_FLAG_SNAPSHOT_EXISTS)); | |
1386 | continue; | |
1387 | } | |
1388 | ASSERT(dbn->dbn_phys.zbm_flags & ZBM_FLAG_SNAPSHOT_EXISTS); | |
1389 | dbn->dbn_phys.zbm_flags &= ~ZBM_FLAG_SNAPSHOT_EXISTS; | |
1390 | VERIFY0(zap_update(dp->dp_meta_objset, head->ds_bookmarks_obj, | |
1391 | dbn->dbn_name, sizeof (uint64_t), | |
1392 | sizeof (zfs_bookmark_phys_t) / sizeof (uint64_t), | |
1393 | &dbn->dbn_phys, tx)); | |
1394 | rv = B_TRUE; | |
1395 | } | |
1396 | dsl_dataset_rele(head, FTAG); | |
1397 | return (rv); | |
1398 | } | |
1399 | ||
1400 | /* | |
1401 | * A snapshot is being created of this (head) dataset. | |
1402 | * | |
1403 | * We don't keep keys in the deadlist for the most recent snapshot, or any | |
1404 | * bookmarks at or after it, because there can't be any blocks on the | |
1405 | * deadlist in this range. Now that the most recent snapshot is after | |
1406 | * all bookmarks, we need to add these keys. Note that the caller always | |
1407 | * adds a key at the previous snapshot, so we only add keys for bookmarks | |
1408 | * after that. | |
1409 | */ | |
1410 | void | |
1411 | dsl_bookmark_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx) | |
1412 | { | |
1413 | uint64_t last_key_added = UINT64_MAX; | |
1414 | for (dsl_bookmark_node_t *dbn = avl_last(&ds->ds_bookmarks); | |
1415 | dbn != NULL && dbn->dbn_phys.zbm_creation_txg > | |
1416 | dsl_dataset_phys(ds)->ds_prev_snap_txg; | |
1417 | dbn = AVL_PREV(&ds->ds_bookmarks, dbn)) { | |
1418 | uint64_t creation_txg = dbn->dbn_phys.zbm_creation_txg; | |
1419 | ASSERT3U(creation_txg, <=, last_key_added); | |
1420 | /* | |
1421 | * Note, there may be multiple bookmarks at this TXG, | |
1422 | * and we only want to add the key for this TXG once. | |
1423 | * The ds_bookmarks AVL is sorted by TXG, so we will visit | |
1424 | * these bookmarks in sequence. | |
1425 | */ | |
1426 | if ((dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN) && | |
1427 | creation_txg != last_key_added) { | |
1428 | dsl_deadlist_add_key(&ds->ds_deadlist, | |
1429 | creation_txg, tx); | |
1430 | last_key_added = creation_txg; | |
1431 | } | |
1432 | } | |
1433 | } | |
1434 | ||
1435 | /* | |
1436 | * The next snapshot of the origin dataset has changed, due to | |
1437 | * promote or clone swap. If there are any bookmarks at this dataset, | |
1438 | * we need to update their zbm_*_freed_before_next_snap to reflect this. | |
1439 | * The head dataset has the relevant bookmarks in ds_bookmarks. | |
1440 | */ | |
1441 | void | |
1442 | dsl_bookmark_next_changed(dsl_dataset_t *head, dsl_dataset_t *origin, | |
1443 | dmu_tx_t *tx) | |
1444 | { | |
1445 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1446 | ||
1447 | /* | |
1448 | * Find the first bookmark that HAS_FBN at the origin snapshot. | |
1449 | */ | |
1450 | dsl_bookmark_node_t search = { 0 }; | |
1451 | avl_index_t idx; | |
1452 | search.dbn_phys.zbm_creation_txg = | |
1453 | dsl_dataset_phys(origin)->ds_creation_txg; | |
1454 | search.dbn_phys.zbm_flags = ZBM_FLAG_HAS_FBN; | |
1455 | /* | |
1456 | * The empty-string name can't be in the AVL, and it compares | |
1457 | * before any entries with this TXG. | |
1458 | */ | |
a926aab9 | 1459 | search.dbn_name = (char *)""; |
30af21b0 PD |
1460 | VERIFY3P(avl_find(&head->ds_bookmarks, &search, &idx), ==, NULL); |
1461 | dsl_bookmark_node_t *dbn = | |
1462 | avl_nearest(&head->ds_bookmarks, idx, AVL_AFTER); | |
1463 | ||
1464 | /* | |
1465 | * Iterate over all bookmarks that are at the origin txg. | |
1466 | * Adjust their FBN based on their new next snapshot. | |
1467 | */ | |
1468 | for (; dbn != NULL && dbn->dbn_phys.zbm_creation_txg == | |
1469 | dsl_dataset_phys(origin)->ds_creation_txg && | |
1470 | (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN); | |
1471 | dbn = AVL_NEXT(&head->ds_bookmarks, dbn)) { | |
1472 | ||
1473 | /* | |
1474 | * Bookmark is at the origin, therefore its | |
1475 | * "next dataset" is changing, so we need | |
1476 | * to reset its FBN by recomputing it in | |
1477 | * dsl_bookmark_set_phys(). | |
1478 | */ | |
1479 | ASSERT3U(dbn->dbn_phys.zbm_guid, ==, | |
1480 | dsl_dataset_phys(origin)->ds_guid); | |
1481 | ASSERT3U(dbn->dbn_phys.zbm_referenced_bytes_refd, ==, | |
1482 | dsl_dataset_phys(origin)->ds_referenced_bytes); | |
1483 | ASSERT(dbn->dbn_phys.zbm_flags & | |
1484 | ZBM_FLAG_SNAPSHOT_EXISTS); | |
1485 | /* | |
1486 | * Save and restore the zbm_redaction_obj, which | |
1487 | * is zeroed by dsl_bookmark_set_phys(). | |
1488 | */ | |
1489 | uint64_t redaction_obj = | |
1490 | dbn->dbn_phys.zbm_redaction_obj; | |
1491 | dsl_bookmark_set_phys(&dbn->dbn_phys, origin); | |
1492 | dbn->dbn_phys.zbm_redaction_obj = redaction_obj; | |
1493 | ||
1494 | VERIFY0(zap_update(dp->dp_meta_objset, head->ds_bookmarks_obj, | |
1495 | dbn->dbn_name, sizeof (uint64_t), | |
1496 | sizeof (zfs_bookmark_phys_t) / sizeof (uint64_t), | |
1497 | &dbn->dbn_phys, tx)); | |
1498 | } | |
1499 | } | |
1500 | ||
1501 | /* | |
1502 | * This block is no longer referenced by this (head) dataset. | |
1503 | * | |
1504 | * Adjust the FBN of any bookmarks that reference this block, whose "next" | |
1505 | * is the head dataset. | |
1506 | */ | |
30af21b0 PD |
1507 | void |
1508 | dsl_bookmark_block_killed(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) | |
1509 | { | |
14e4e3cb AZ |
1510 | (void) tx; |
1511 | ||
30af21b0 PD |
1512 | /* |
1513 | * Iterate over bookmarks whose "next" is the head dataset. | |
1514 | */ | |
1515 | for (dsl_bookmark_node_t *dbn = avl_last(&ds->ds_bookmarks); | |
1516 | dbn != NULL && dbn->dbn_phys.zbm_creation_txg >= | |
1517 | dsl_dataset_phys(ds)->ds_prev_snap_txg; | |
1518 | dbn = AVL_PREV(&ds->ds_bookmarks, dbn)) { | |
1519 | /* | |
1520 | * If the block was live (referenced) at the time of this | |
1521 | * bookmark, add its space to the bookmark's FBN. | |
1522 | */ | |
1523 | if (bp->blk_birth <= dbn->dbn_phys.zbm_creation_txg && | |
1524 | (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)) { | |
1525 | mutex_enter(&dbn->dbn_lock); | |
1526 | dbn->dbn_phys.zbm_referenced_freed_before_next_snap += | |
1527 | bp_get_dsize_sync(dsl_dataset_get_spa(ds), bp); | |
1528 | dbn->dbn_phys.zbm_compressed_freed_before_next_snap += | |
1529 | BP_GET_PSIZE(bp); | |
1530 | dbn->dbn_phys.zbm_uncompressed_freed_before_next_snap += | |
1531 | BP_GET_UCSIZE(bp); | |
1532 | /* | |
1533 | * Changing the ZAP object here would be too | |
1534 | * expensive. Also, we may be called from the zio | |
1535 | * interrupt thread, which can't block on i/o. | |
1536 | * Therefore, we mark this bookmark as dirty and | |
1537 | * modify the ZAP once per txg, in | |
1538 | * dsl_bookmark_sync_done(). | |
1539 | */ | |
1540 | dbn->dbn_dirty = B_TRUE; | |
1541 | mutex_exit(&dbn->dbn_lock); | |
1542 | } | |
1543 | } | |
1544 | } | |
1545 | ||
1546 | void | |
1547 | dsl_bookmark_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx) | |
1548 | { | |
1549 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1550 | ||
1551 | if (dsl_dataset_is_snapshot(ds)) | |
1552 | return; | |
1553 | ||
1554 | /* | |
1555 | * We only dirty bookmarks that are at or after the most recent | |
1556 | * snapshot. We can't create snapshots between | |
1557 | * dsl_bookmark_block_killed() and dsl_bookmark_sync_done(), so we | |
1558 | * don't need to look at any bookmarks before ds_prev_snap_txg. | |
1559 | */ | |
1560 | for (dsl_bookmark_node_t *dbn = avl_last(&ds->ds_bookmarks); | |
1561 | dbn != NULL && dbn->dbn_phys.zbm_creation_txg >= | |
1562 | dsl_dataset_phys(ds)->ds_prev_snap_txg; | |
1563 | dbn = AVL_PREV(&ds->ds_bookmarks, dbn)) { | |
1564 | if (dbn->dbn_dirty) { | |
1565 | /* | |
1566 | * We only dirty nodes with HAS_FBN, therefore | |
1567 | * we can always use the current bookmark struct size. | |
1568 | */ | |
1569 | ASSERT(dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN); | |
1570 | VERIFY0(zap_update(dp->dp_meta_objset, | |
1571 | ds->ds_bookmarks_obj, | |
1572 | dbn->dbn_name, sizeof (uint64_t), | |
1573 | sizeof (zfs_bookmark_phys_t) / sizeof (uint64_t), | |
1574 | &dbn->dbn_phys, tx)); | |
1575 | dbn->dbn_dirty = B_FALSE; | |
1576 | } | |
1577 | } | |
1578 | #ifdef ZFS_DEBUG | |
1579 | for (dsl_bookmark_node_t *dbn = avl_first(&ds->ds_bookmarks); | |
1580 | dbn != NULL; dbn = AVL_NEXT(&ds->ds_bookmarks, dbn)) { | |
1581 | ASSERT(!dbn->dbn_dirty); | |
1582 | } | |
1583 | #endif | |
1584 | } | |
1585 | ||
1586 | /* | |
1587 | * Return the TXG of the most recent bookmark (or 0 if there are no bookmarks). | |
1588 | */ | |
1589 | uint64_t | |
1590 | dsl_bookmark_latest_txg(dsl_dataset_t *ds) | |
1591 | { | |
1592 | ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); | |
1593 | dsl_bookmark_node_t *dbn = avl_last(&ds->ds_bookmarks); | |
1594 | if (dbn == NULL) | |
1595 | return (0); | |
1596 | return (dbn->dbn_phys.zbm_creation_txg); | |
1597 | } | |
1598 | ||
30af21b0 PD |
1599 | /* |
1600 | * Compare the redact_block_phys_t to the bookmark. If the last block in the | |
1601 | * redact_block_phys_t is before the bookmark, return -1. If the first block in | |
1602 | * the redact_block_phys_t is after the bookmark, return 1. Otherwise, the | |
1603 | * bookmark is inside the range of the redact_block_phys_t, and we return 0. | |
1604 | */ | |
1605 | static int | |
1606 | redact_block_zb_compare(redact_block_phys_t *first, | |
1607 | zbookmark_phys_t *second) | |
1608 | { | |
1609 | /* | |
1610 | * If the block_phys is for a previous object, or the last block in the | |
1611 | * block_phys is strictly before the block in the bookmark, the | |
1612 | * block_phys is earlier. | |
1613 | */ | |
1614 | if (first->rbp_object < second->zb_object || | |
1615 | (first->rbp_object == second->zb_object && | |
1616 | first->rbp_blkid + (redact_block_get_count(first) - 1) < | |
1617 | second->zb_blkid)) { | |
1618 | return (-1); | |
1619 | } | |
1620 | ||
1621 | /* | |
1622 | * If the bookmark is for a previous object, or the block in the | |
1623 | * bookmark is strictly before the first block in the block_phys, the | |
1624 | * bookmark is earlier. | |
1625 | */ | |
1626 | if (first->rbp_object > second->zb_object || | |
1627 | (first->rbp_object == second->zb_object && | |
1628 | first->rbp_blkid > second->zb_blkid)) { | |
1629 | return (1); | |
1630 | } | |
1631 | ||
1632 | return (0); | |
1633 | } | |
1634 | ||
1635 | /* | |
1636 | * Traverse the redaction list in the provided object, and call the callback for | |
1637 | * each entry we find. Don't call the callback for any records before resume. | |
1638 | */ | |
1639 | int | |
1640 | dsl_redaction_list_traverse(redaction_list_t *rl, zbookmark_phys_t *resume, | |
1641 | rl_traverse_callback_t cb, void *arg) | |
1642 | { | |
1643 | objset_t *mos = rl->rl_mos; | |
30af21b0 PD |
1644 | int err = 0; |
1645 | ||
1646 | if (rl->rl_phys->rlp_last_object != UINT64_MAX || | |
1647 | rl->rl_phys->rlp_last_blkid != UINT64_MAX) { | |
1648 | /* | |
1649 | * When we finish a send, we update the last object and offset | |
1650 | * to UINT64_MAX. If a send fails partway through, the last | |
1651 | * object and offset will have some other value, indicating how | |
1652 | * far the send got. The redaction list must be complete before | |
1653 | * it can be traversed, so return EINVAL if the last object and | |
1654 | * blkid are not set to UINT64_MAX. | |
1655 | */ | |
1656 | return (SET_ERROR(EINVAL)); | |
1657 | } | |
1658 | ||
1659 | /* | |
7d4b365c PD |
1660 | * This allows us to skip the binary search and resume checking logic |
1661 | * below, if we're not resuming a redacted send. | |
1662 | */ | |
1663 | if (ZB_IS_ZERO(resume)) | |
1664 | resume = NULL; | |
1665 | ||
1666 | /* | |
1667 | * Binary search for the point to resume from. | |
30af21b0 | 1668 | */ |
7d4b365c PD |
1669 | uint64_t maxidx = rl->rl_phys->rlp_num_entries - 1; |
1670 | uint64_t minidx = 0; | |
1671 | while (resume != NULL && maxidx > minidx) { | |
1672 | redact_block_phys_t rbp = { 0 }; | |
1673 | ASSERT3U(maxidx, >, minidx); | |
1674 | uint64_t mididx = minidx + ((maxidx - minidx) / 2); | |
1675 | err = dmu_read(mos, rl->rl_object, mididx * sizeof (rbp), | |
1676 | sizeof (rbp), &rbp, DMU_READ_NO_PREFETCH); | |
30af21b0 PD |
1677 | if (err != 0) |
1678 | break; | |
1679 | ||
7d4b365c | 1680 | int cmp = redact_block_zb_compare(&rbp, resume); |
30af21b0 | 1681 | |
7d4b365c PD |
1682 | if (cmp == 0) { |
1683 | minidx = mididx; | |
30af21b0 | 1684 | break; |
7d4b365c PD |
1685 | } else if (cmp > 0) { |
1686 | maxidx = | |
1687 | (mididx == minidx ? minidx : mididx - 1); | |
1688 | } else { | |
1689 | minidx = mididx + 1; | |
1690 | } | |
30af21b0 PD |
1691 | } |
1692 | ||
7d4b365c PD |
1693 | unsigned int bufsize = SPA_OLD_MAXBLOCKSIZE; |
1694 | redact_block_phys_t *buf = zio_data_buf_alloc(bufsize); | |
1695 | ||
1696 | unsigned int entries_per_buf = bufsize / sizeof (redact_block_phys_t); | |
1697 | uint64_t start_block = minidx / entries_per_buf; | |
1698 | err = dmu_read(mos, rl->rl_object, start_block * bufsize, bufsize, buf, | |
1699 | DMU_READ_PREFETCH); | |
1700 | ||
1701 | for (uint64_t curidx = minidx; | |
30af21b0 PD |
1702 | err == 0 && curidx < rl->rl_phys->rlp_num_entries; |
1703 | curidx++) { | |
1704 | /* | |
1705 | * We read in the redaction list one block at a time. Once we | |
1706 | * finish with all the entries in a given block, we read in a | |
1707 | * new one. The predictive prefetcher will take care of any | |
1708 | * prefetching, and this code shouldn't be the bottleneck, so we | |
1709 | * don't need to do manual prefetching. | |
1710 | */ | |
7d4b365c | 1711 | if (curidx % entries_per_buf == 0) { |
30af21b0 PD |
1712 | err = dmu_read(mos, rl->rl_object, curidx * |
1713 | sizeof (*buf), bufsize, buf, | |
1714 | DMU_READ_PREFETCH); | |
1715 | if (err != 0) | |
1716 | break; | |
1717 | } | |
7d4b365c | 1718 | redact_block_phys_t *rb = &buf[curidx % entries_per_buf]; |
30af21b0 PD |
1719 | /* |
1720 | * If resume is non-null, we should either not send the data, or | |
1721 | * null out resume so we don't have to keep doing these | |
1722 | * comparisons. | |
1723 | */ | |
1724 | if (resume != NULL) { | |
7d4b365c PD |
1725 | /* |
1726 | * It is possible that after the binary search we got | |
1727 | * a record before the resume point. There's two cases | |
1728 | * where this can occur. If the record is the last | |
1729 | * redaction record, and the resume point is after the | |
1730 | * end of the redacted data, curidx will be the last | |
1731 | * redaction record. In that case, the loop will end | |
1732 | * after this iteration. The second case is if the | |
1733 | * resume point is between two redaction records, the | |
1734 | * binary search can return either the record before | |
1735 | * or after the resume point. In that case, the next | |
1736 | * iteration will be greater than the resume point. | |
1737 | */ | |
30af21b0 | 1738 | if (redact_block_zb_compare(rb, resume) < 0) { |
7d4b365c | 1739 | ASSERT3U(curidx, ==, minidx); |
30af21b0 PD |
1740 | continue; |
1741 | } else { | |
1742 | /* | |
1743 | * If the place to resume is in the middle of | |
1744 | * the range described by this | |
1745 | * redact_block_phys, then modify the | |
1746 | * redact_block_phys in memory so we generate | |
1747 | * the right records. | |
1748 | */ | |
1749 | if (resume->zb_object == rb->rbp_object && | |
1750 | resume->zb_blkid > rb->rbp_blkid) { | |
1751 | uint64_t diff = resume->zb_blkid - | |
1752 | rb->rbp_blkid; | |
1753 | rb->rbp_blkid = resume->zb_blkid; | |
1754 | redact_block_set_count(rb, | |
1755 | redact_block_get_count(rb) - diff); | |
1756 | } | |
1757 | resume = NULL; | |
1758 | } | |
1759 | } | |
1760 | ||
7d4b365c PD |
1761 | if (cb(rb, arg) != 0) { |
1762 | err = EINTR; | |
30af21b0 | 1763 | break; |
7d4b365c | 1764 | } |
30af21b0 PD |
1765 | } |
1766 | ||
1767 | zio_data_buf_free(buf, bufsize); | |
1768 | return (err); | |
1769 | } |