xref: /freebsd/usr.sbin/makefs/zfs/dsl.c (revision 59d6422d6f21fd5cf4709ce9fcada54d3925a4f6)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2022 The FreeBSD Foundation
5  *
6  * This software was developed by Mark Johnston under sponsorship from
7  * the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions are
11  * met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in
16  *    the documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/param.h>
32 #include <assert.h>
33 #include <stdlib.h>
34 #include <string.h>
35 
36 #include <util.h>
37 
38 #include "makefs.h"
39 #include "zfs.h"
40 
41 typedef struct zfs_dsl_dataset {
42 	zfs_objset_t	*os;		/* referenced objset, may be null */
43 	dsl_dataset_phys_t *phys;	/* on-disk representation */
44 	uint64_t	dsid;		/* DSL dataset dnode */
45 
46 	struct zfs_dsl_dir *dir;	/* containing parent */
47 } zfs_dsl_dataset_t;
48 
49 typedef STAILQ_HEAD(zfs_dsl_dir_list, zfs_dsl_dir) zfs_dsl_dir_list_t;
50 
51 typedef struct zfs_dsl_dir {
52 	char		*fullname;	/* full dataset name */
53 	char		*name;		/* basename(fullname) */
54 	dsl_dir_phys_t	*phys;		/* on-disk representation */
55 	nvlist_t	*propsnv;	/* properties saved in propszap */
56 
57 	zfs_dsl_dataset_t *headds;	/* principal dataset, may be null */
58 
59 	uint64_t	dirid;		/* DSL directory dnode */
60 	zfs_zap_t	*propszap;	/* dataset properties */
61 	zfs_zap_t	*childzap;	/* child directories */
62 
63 	/* DSL directory tree linkage. */
64 	struct zfs_dsl_dir *parent;
65 	zfs_dsl_dir_list_t children;
66 	STAILQ_ENTRY(zfs_dsl_dir) next;
67 } zfs_dsl_dir_t;
68 
69 static zfs_dsl_dir_t *dsl_dir_alloc(zfs_opt_t *zfs, const char *name);
70 static zfs_dsl_dataset_t *dsl_dataset_alloc(zfs_opt_t *zfs, zfs_dsl_dir_t *dir);
71 
72 static int
nvlist_find_string(nvlist_t * nvl,const char * key,char ** retp)73 nvlist_find_string(nvlist_t *nvl, const char *key, char **retp)
74 {
75 	char *str;
76 	int error, len;
77 
78 	error = nvlist_find(nvl, key, DATA_TYPE_STRING, NULL, &str, &len);
79 	if (error == 0) {
80 		*retp = ecalloc(1, len + 1);
81 		memcpy(*retp, str, len);
82 	}
83 	return (error);
84 }
85 
86 static int
nvlist_find_uint64(nvlist_t * nvl,const char * key,uint64_t * retp)87 nvlist_find_uint64(nvlist_t *nvl, const char *key, uint64_t *retp)
88 {
89 	return (nvlist_find(nvl, key, DATA_TYPE_UINT64, NULL, retp, NULL));
90 }
91 
92 /*
93  * Return an allocated string containing the head dataset's mountpoint,
94  * including the root path prefix.
95  *
96  * If the dataset has a mountpoint property, it is returned.  Otherwise we have
97  * to follow ZFS' inheritance rules.
98  */
99 char *
dsl_dir_get_mountpoint(zfs_opt_t * zfs,zfs_dsl_dir_t * dir)100 dsl_dir_get_mountpoint(zfs_opt_t *zfs, zfs_dsl_dir_t *dir)
101 {
102 	zfs_dsl_dir_t *pdir;
103 	char *mountpoint;
104 
105 	if (nvlist_find_string(dir->propsnv, "mountpoint", &mountpoint) == 0) {
106 		if (strcmp(mountpoint, "none") == 0 ||
107 		    strcmp(mountpoint, "legacy") == 0)
108 			return (NULL);
109 	} else {
110 		/*
111 		 * If we don't have a mountpoint, it's inherited from one of our
112 		 * ancestors.  Walk up the hierarchy until we find it, building
113 		 * up our mountpoint along the way.  The mountpoint property is
114 		 * always set for the root dataset.
115 		 */
116 		for (pdir = dir->parent, mountpoint = estrdup(dir->name);;
117 		    pdir = pdir->parent) {
118 			char *origmountpoint, *tmp;
119 
120 			origmountpoint = mountpoint;
121 
122 			if (nvlist_find_string(pdir->propsnv, "mountpoint",
123 			    &tmp) == 0) {
124 				(void)easprintf(&mountpoint, "%s%s%s", tmp,
125 				    tmp[strlen(tmp) - 1] == '/' ?  "" : "/",
126 				    origmountpoint);
127 				free(tmp);
128 				free(origmountpoint);
129 				break;
130 			}
131 
132 			(void)easprintf(&mountpoint, "%s/%s", pdir->name,
133 			    origmountpoint);
134 			free(origmountpoint);
135 		}
136 	}
137 	assert(mountpoint[0] == '/');
138 	assert(strstr(mountpoint, zfs->rootpath) == mountpoint);
139 
140 	return (mountpoint);
141 }
142 
143 int
dsl_dir_get_canmount(zfs_dsl_dir_t * dir,uint64_t * canmountp)144 dsl_dir_get_canmount(zfs_dsl_dir_t *dir, uint64_t *canmountp)
145 {
146 	return (nvlist_find_uint64(dir->propsnv, "canmount", canmountp));
147 }
148 
149 /*
150  * Handle dataset properties that we know about; stash them into an nvlist to be
151  * written later to the properties ZAP object.
152  *
153  * If the set of properties we handle grows too much, we should probably explore
154  * using libzfs to manage them.
155  */
156 static void
dsl_dir_set_prop(zfs_opt_t * zfs,zfs_dsl_dir_t * dir,const char * key,const char * val)157 dsl_dir_set_prop(zfs_opt_t *zfs, zfs_dsl_dir_t *dir, const char *key,
158     const char *val)
159 {
160 	nvlist_t *nvl;
161 
162 	nvl = dir->propsnv;
163 	if (val == NULL || val[0] == '\0')
164 		errx(1, "missing value for property `%s'", key);
165 	if (nvpair_find(nvl, key) != NULL)
166 		errx(1, "property `%s' already set", key);
167 
168 	if (strcmp(key, "mountpoint") == 0) {
169 		if (strcmp(val, "none") != 0 && strcmp(val, "legacy") != 0) {
170 			if (val[0] != '/')
171 				errx(1, "mountpoint `%s' is not absolute", val);
172 			if (strcmp(val, zfs->rootpath) != 0 &&
173 			    strcmp(zfs->rootpath, "/") != 0 &&
174 			    (strstr(val, zfs->rootpath) != val ||
175 			     val[strlen(zfs->rootpath)] != '/')) {
176 				errx(1, "mountpoint `%s' is not prefixed by "
177 				    "the root path `%s'", val, zfs->rootpath);
178 			}
179 		}
180 		(void)nvlist_add_string(nvl, key, val);
181 	} else if (strcmp(key, "atime") == 0 || strcmp(key, "exec") == 0 ||
182 	    strcmp(key, "setuid") == 0) {
183 		if (strcmp(val, "on") == 0)
184 			(void)nvlist_add_uint64(nvl, key, 1);
185 		else if (strcmp(val, "off") == 0)
186 			(void)nvlist_add_uint64(nvl, key, 0);
187 		else
188 			errx(1, "invalid value `%s' for %s", val, key);
189 	} else if (strcmp(key, "canmount") == 0) {
190 		if (strcmp(val, "noauto") == 0)
191 			(void)nvlist_add_uint64(nvl, key, 2);
192 		else if (strcmp(val, "on") == 0)
193 			(void)nvlist_add_uint64(nvl, key, 1);
194 		else if (strcmp(val, "off") == 0)
195 			(void)nvlist_add_uint64(nvl, key, 0);
196 		else
197 			errx(1, "invalid value `%s' for %s", val, key);
198 	} else if (strcmp(key, "compression") == 0) {
199 		size_t i;
200 
201 		const struct zfs_compression_algorithm {
202 			const char *name;
203 			enum zio_compress alg;
204 		} compression_algorithms[] = {
205 			{ "off", ZIO_COMPRESS_OFF },
206 			{ "on", ZIO_COMPRESS_ON },
207 			{ "lzjb", ZIO_COMPRESS_LZJB },
208 			{ "gzip", ZIO_COMPRESS_GZIP_6 },
209 			{ "gzip-1", ZIO_COMPRESS_GZIP_1 },
210 			{ "gzip-2", ZIO_COMPRESS_GZIP_2 },
211 			{ "gzip-3", ZIO_COMPRESS_GZIP_3 },
212 			{ "gzip-4", ZIO_COMPRESS_GZIP_4 },
213 			{ "gzip-5", ZIO_COMPRESS_GZIP_5 },
214 			{ "gzip-6", ZIO_COMPRESS_GZIP_6 },
215 			{ "gzip-7", ZIO_COMPRESS_GZIP_7 },
216 			{ "gzip-8", ZIO_COMPRESS_GZIP_8 },
217 			{ "gzip-9", ZIO_COMPRESS_GZIP_9 },
218 			{ "zle", ZIO_COMPRESS_ZLE },
219 			{ "lz4", ZIO_COMPRESS_LZ4 },
220 			{ "zstd", ZIO_COMPRESS_ZSTD },
221 		};
222 		for (i = 0; i < nitems(compression_algorithms); i++) {
223 			if (strcmp(val, compression_algorithms[i].name) == 0) {
224 				nvlist_add_uint64(nvl, key,
225 				    compression_algorithms[i].alg);
226 				break;
227 			}
228 		}
229 		if (i == nitems(compression_algorithms))
230 			errx(1, "invalid compression algorithm `%s'", val);
231 	} else {
232 		errx(1, "unknown property `%s'", key);
233 	}
234 }
235 
236 static zfs_dsl_dir_t *
dsl_metadir_alloc(zfs_opt_t * zfs,const char * name)237 dsl_metadir_alloc(zfs_opt_t *zfs, const char *name)
238 {
239 	zfs_dsl_dir_t *dir;
240 	char *path;
241 
242 	(void)easprintf(&path, "%s/%s", zfs->poolname, name);
243 	dir = dsl_dir_alloc(zfs, path);
244 	free(path);
245 	return (dir);
246 }
247 
248 static void
dsl_origindir_init(zfs_opt_t * zfs)249 dsl_origindir_init(zfs_opt_t *zfs)
250 {
251 	dnode_phys_t *clones;
252 	uint64_t clonesid;
253 
254 	zfs->origindsldir = dsl_metadir_alloc(zfs, "$ORIGIN");
255 	zfs->originds = dsl_dataset_alloc(zfs, zfs->origindsldir);
256 	zfs->snapds = dsl_dataset_alloc(zfs, zfs->origindsldir);
257 
258 	clones = objset_dnode_alloc(zfs->mos, DMU_OT_DSL_CLONES, &clonesid);
259 	zfs->cloneszap = zap_alloc(zfs->mos, clones);
260 	zfs->origindsldir->phys->dd_clones = clonesid;
261 }
262 
263 void
dsl_init(zfs_opt_t * zfs)264 dsl_init(zfs_opt_t *zfs)
265 {
266 	zfs_dsl_dir_t *dir;
267 	struct dataset_desc *d;
268 	const char *dspropdelim;
269 
270 	dspropdelim = ";";
271 
272 	zfs->rootdsldir = dsl_dir_alloc(zfs, NULL);
273 
274 	zfs->rootds = dsl_dataset_alloc(zfs, zfs->rootdsldir);
275 	zfs->rootdsldir->headds = zfs->rootds;
276 
277 	zfs->mosdsldir = dsl_metadir_alloc(zfs, "$MOS");
278 	zfs->freedsldir = dsl_metadir_alloc(zfs, "$FREE");
279 	dsl_origindir_init(zfs);
280 
281 	/*
282 	 * Go through the list of user-specified datasets and create DSL objects
283 	 * for them.
284 	 */
285 	STAILQ_FOREACH(d, &zfs->datasetdescs, next) {
286 		char *dsname, *next, *params, *param, *nextparam;
287 
288 		params = d->params;
289 		dsname = strsep(&params, dspropdelim);
290 
291 		if (strcmp(dsname, zfs->poolname) == 0) {
292 			/*
293 			 * This is the root dataset; it's already created, so
294 			 * we're just setting options.
295 			 */
296 			dir = zfs->rootdsldir;
297 		} else {
298 			/*
299 			 * This dataset must be a child of the root dataset.
300 			 */
301 			if (strstr(dsname, zfs->poolname) != dsname ||
302 			    (next = strchr(dsname, '/')) == NULL ||
303 			    (size_t)(next - dsname) != strlen(zfs->poolname)) {
304 				errx(1, "dataset `%s' must be a child of `%s'",
305 				    dsname, zfs->poolname);
306 			}
307 			dir = dsl_dir_alloc(zfs, dsname);
308 			dir->headds = dsl_dataset_alloc(zfs, dir);
309 		}
310 
311 		for (nextparam = param = params; nextparam != NULL;) {
312 			char *key, *val;
313 
314 			param = strsep(&nextparam, dspropdelim);
315 
316 			key = val = param;
317 			key = strsep(&val, "=");
318 			dsl_dir_set_prop(zfs, dir, key, val);
319 		}
320 	}
321 
322 	/*
323 	 * Set the root dataset's mount point and compression strategy if the
324 	 * user didn't override the defaults.
325 	 */
326 	if (nvpair_find(zfs->rootdsldir->propsnv, "compression") == NULL) {
327 		(void)nvlist_add_uint64(zfs->rootdsldir->propsnv,
328 		    "compression", ZIO_COMPRESS_OFF);
329 	}
330 	if (nvpair_find(zfs->rootdsldir->propsnv, "mountpoint") == NULL) {
331 		(void)nvlist_add_string(zfs->rootdsldir->propsnv, "mountpoint",
332 		    zfs->rootpath);
333 	}
334 }
335 
336 uint64_t
dsl_dir_id(zfs_dsl_dir_t * dir)337 dsl_dir_id(zfs_dsl_dir_t *dir)
338 {
339 	return (dir->dirid);
340 }
341 
342 uint64_t
dsl_dir_dataset_id(zfs_dsl_dir_t * dir)343 dsl_dir_dataset_id(zfs_dsl_dir_t *dir)
344 {
345 	return (dir->headds->dsid);
346 }
347 
348 static void
dsl_dir_foreach_post(zfs_opt_t * zfs,zfs_dsl_dir_t * dsldir,void (* cb)(zfs_opt_t *,zfs_dsl_dir_t *,void *),void * arg)349 dsl_dir_foreach_post(zfs_opt_t *zfs, zfs_dsl_dir_t *dsldir,
350     void (*cb)(zfs_opt_t *, zfs_dsl_dir_t *, void *), void *arg)
351 {
352 	zfs_dsl_dir_t *cdsldir;
353 
354 	STAILQ_FOREACH(cdsldir, &dsldir->children, next) {
355 		dsl_dir_foreach_post(zfs, cdsldir, cb, arg);
356 	}
357 	cb(zfs, dsldir, arg);
358 }
359 
360 /*
361  * Used when the caller doesn't care about the order one way or another.
362  */
363 void
dsl_dir_foreach(zfs_opt_t * zfs,zfs_dsl_dir_t * dsldir,void (* cb)(zfs_opt_t *,zfs_dsl_dir_t *,void *),void * arg)364 dsl_dir_foreach(zfs_opt_t *zfs, zfs_dsl_dir_t *dsldir,
365     void (*cb)(zfs_opt_t *, zfs_dsl_dir_t *, void *), void *arg)
366 {
367 	dsl_dir_foreach_post(zfs, dsldir, cb, arg);
368 }
369 
370 const char *
dsl_dir_fullname(const zfs_dsl_dir_t * dir)371 dsl_dir_fullname(const zfs_dsl_dir_t *dir)
372 {
373 	return (dir->fullname);
374 }
375 
376 /*
377  * Create a DSL directory, which is effectively an entry in the ZFS namespace.
378  * We always create a root DSL directory, whose name is the pool's name, and
379  * several metadata directories.
380  *
381  * Each directory has two ZAP objects, one pointing to child directories, and
382  * one for properties (which are inherited by children unless overridden).
383  * Directories typically reference a DSL dataset, the "head dataset", which
384  * points to an object set.
385  */
386 static zfs_dsl_dir_t *
dsl_dir_alloc(zfs_opt_t * zfs,const char * name)387 dsl_dir_alloc(zfs_opt_t *zfs, const char *name)
388 {
389 	zfs_dsl_dir_list_t l, *lp;
390 	zfs_dsl_dir_t *dir, *parent;
391 	dnode_phys_t *dnode;
392 	char *dirname, *nextdir, *origname;
393 	uint64_t childid, propsid;
394 
395 	dir = ecalloc(1, sizeof(*dir));
396 
397 	dnode = objset_dnode_bonus_alloc(zfs->mos, DMU_OT_DSL_DIR,
398 	    DMU_OT_DSL_DIR, sizeof(dsl_dir_phys_t), &dir->dirid);
399 	dir->phys = (dsl_dir_phys_t *)DN_BONUS(dnode);
400 
401 	dnode = objset_dnode_alloc(zfs->mos, DMU_OT_DSL_PROPS, &propsid);
402 	dir->propszap = zap_alloc(zfs->mos, dnode);
403 
404 	dnode = objset_dnode_alloc(zfs->mos, DMU_OT_DSL_DIR_CHILD_MAP,
405 	    &childid);
406 	dir->childzap = zap_alloc(zfs->mos, dnode);
407 
408 	dir->propsnv = nvlist_create(NV_UNIQUE_NAME);
409 	STAILQ_INIT(&dir->children);
410 
411 	dir->phys->dd_child_dir_zapobj = childid;
412 	dir->phys->dd_props_zapobj = propsid;
413 
414 	if (name == NULL) {
415 		/*
416 		 * This is the root DSL directory.
417 		 */
418 		dir->name = estrdup(zfs->poolname);
419 		dir->fullname = estrdup(zfs->poolname);
420 		dir->parent = NULL;
421 		dir->phys->dd_parent_obj = 0;
422 
423 		assert(zfs->rootdsldir == NULL);
424 		zfs->rootdsldir = dir;
425 		return (dir);
426 	}
427 
428 	/*
429 	 * Insert the new directory into the hierarchy.  Currently this must be
430 	 * done in order, e.g., when creating pool/a/b, pool/a must already
431 	 * exist.
432 	 */
433 	STAILQ_INIT(&l);
434 	STAILQ_INSERT_HEAD(&l, zfs->rootdsldir, next);
435 	origname = dirname = nextdir = estrdup(name);
436 	parent = NULL;
437 	for (lp = &l;; lp = &parent->children) {
438 		dirname = strsep(&nextdir, "/");
439 		if (nextdir == NULL)
440 			break;
441 
442 		STAILQ_FOREACH(parent, lp, next) {
443 			if (strcmp(parent->name, dirname) == 0)
444 				break;
445 		}
446 		if (parent == NULL) {
447 			errx(1, "no parent at `%s' for filesystem `%s'",
448 			    dirname, name);
449 		}
450 	}
451 
452 	dir->fullname = estrdup(name);
453 	dir->name = estrdup(dirname);
454 	free(origname);
455 	STAILQ_INSERT_TAIL(lp, dir, next);
456 	zap_add_uint64(parent->childzap, dir->name, dir->dirid);
457 
458 	dir->parent = parent;
459 	dir->phys->dd_parent_obj = parent->dirid;
460 	return (dir);
461 }
462 
463 static void
dsl_dir_size_add(zfs_dsl_dir_t * dir,uint64_t bytes)464 dsl_dir_size_add(zfs_dsl_dir_t *dir, uint64_t bytes)
465 {
466 	dir->phys->dd_used_bytes += bytes;
467 	dir->phys->dd_compressed_bytes += bytes;
468 	dir->phys->dd_uncompressed_bytes += bytes;
469 }
470 
471 /*
472  * See dsl_dir_root_finalize().
473  */
474 void
dsl_dir_root_finalize(zfs_opt_t * zfs,uint64_t bytes)475 dsl_dir_root_finalize(zfs_opt_t *zfs, uint64_t bytes)
476 {
477 	dsl_dir_size_add(zfs->mosdsldir, bytes);
478 	zfs->mosdsldir->phys->dd_used_breakdown[DD_USED_HEAD] += bytes;
479 
480 	dsl_dir_size_add(zfs->rootdsldir, bytes);
481 	zfs->rootdsldir->phys->dd_used_breakdown[DD_USED_CHILD] += bytes;
482 }
483 
484 /*
485  * Convert dataset properties into entries in the DSL directory's properties
486  * ZAP.
487  */
488 static void
dsl_dir_finalize_props(zfs_dsl_dir_t * dir)489 dsl_dir_finalize_props(zfs_dsl_dir_t *dir)
490 {
491 	for (nvp_header_t *nvh = NULL;
492 	    (nvh = nvlist_next_nvpair(dir->propsnv, nvh)) != NULL;) {
493 		nv_string_t *nvname;
494 		nv_pair_data_t *nvdata;
495 		char *name;
496 
497 		nvname = (nv_string_t *)(nvh + 1);
498 		nvdata = (nv_pair_data_t *)(&nvname->nv_data[0] +
499 		    NV_ALIGN4(nvname->nv_size));
500 
501 		name = nvstring_get(nvname);
502 		switch (nvdata->nv_type) {
503 		case DATA_TYPE_UINT64: {
504 			uint64_t val;
505 
506 			memcpy(&val, &nvdata->nv_data[0], sizeof(uint64_t));
507 			zap_add_uint64(dir->propszap, name, val);
508 			break;
509 		}
510 		case DATA_TYPE_STRING: {
511 			nv_string_t *nvstr;
512 			char *val;
513 
514 			nvstr = (nv_string_t *)&nvdata->nv_data[0];
515 			val = nvstring_get(nvstr);
516 			zap_add_string(dir->propszap, name, val);
517 			free(val);
518 			break;
519 		}
520 		default:
521 			assert(0);
522 		}
523 		free(name);
524 	}
525 }
526 
527 static void
dsl_dir_finalize(zfs_opt_t * zfs,zfs_dsl_dir_t * dir,void * arg __unused)528 dsl_dir_finalize(zfs_opt_t *zfs, zfs_dsl_dir_t *dir, void *arg __unused)
529 {
530 	zfs_dsl_dir_t *cdir;
531 	dnode_phys_t *snapnames;
532 	zfs_dsl_dataset_t *headds;
533 	zfs_objset_t *os;
534 	uint64_t bytes, childbytes, snapnamesid;
535 
536 	dsl_dir_finalize_props(dir);
537 	zap_write(zfs, dir->propszap);
538 	zap_write(zfs, dir->childzap);
539 
540 	headds = dir->headds;
541 	if (headds == NULL)
542 		return;
543 	os = headds->os;
544 	if (os == NULL)
545 		return;
546 
547 	snapnames = objset_dnode_alloc(zfs->mos, DMU_OT_DSL_DS_SNAP_MAP,
548 	    &snapnamesid);
549 	zap_write(zfs, zap_alloc(zfs->mos, snapnames));
550 
551 	dir->phys->dd_head_dataset_obj = headds->dsid;
552 	dir->phys->dd_clone_parent_obj = zfs->snapds->dsid;
553 	headds->phys->ds_prev_snap_obj = zfs->snapds->dsid;
554 	headds->phys->ds_snapnames_zapobj = snapnamesid;
555 	objset_root_blkptr_copy(os, &headds->phys->ds_bp);
556 
557 	zfs->snapds->phys->ds_num_children++;
558 	zap_add_uint64_self(zfs->cloneszap, headds->dsid);
559 
560 	bytes = objset_space(os);
561 	headds->phys->ds_used_bytes = bytes;
562 	headds->phys->ds_uncompressed_bytes = bytes;
563 	headds->phys->ds_compressed_bytes = bytes;
564 
565 	childbytes = 0;
566 	STAILQ_FOREACH(cdir, &dir->children, next) {
567 		/*
568 		 * The root directory needs a special case: the amount of
569 		 * space used for the MOS isn't known until everything else is
570 		 * finalized, so it can't be accounted in the MOS directory's
571 		 * parent until then, at which point dsl_dir_root_finalize() is
572 		 * called.
573 		 */
574 		if (dir == zfs->rootdsldir && cdir == zfs->mosdsldir)
575 			continue;
576 		childbytes += cdir->phys->dd_used_bytes;
577 	}
578 	dsl_dir_size_add(dir, bytes + childbytes);
579 
580 	dir->phys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
581 	dir->phys->dd_used_breakdown[DD_USED_HEAD] = bytes;
582 	dir->phys->dd_used_breakdown[DD_USED_CHILD] = childbytes;
583 }
584 
585 void
dsl_write(zfs_opt_t * zfs)586 dsl_write(zfs_opt_t *zfs)
587 {
588 	zfs_zap_t *snapnameszap;
589 	dnode_phys_t *snapnames;
590 	uint64_t snapmapid;
591 
592 	/*
593 	 * Perform accounting, starting from the leaves of the DSL directory
594 	 * tree.  Accounting for $MOS is done later, once we've finished
595 	 * allocating space.
596 	 */
597 	dsl_dir_foreach_post(zfs, zfs->rootdsldir, dsl_dir_finalize, NULL);
598 
599 	snapnames = objset_dnode_alloc(zfs->mos, DMU_OT_DSL_DS_SNAP_MAP,
600 	    &snapmapid);
601 	snapnameszap = zap_alloc(zfs->mos, snapnames);
602 	zap_add_uint64(snapnameszap, "$ORIGIN", zfs->snapds->dsid);
603 	zap_write(zfs, snapnameszap);
604 
605 	zfs->origindsldir->phys->dd_head_dataset_obj = zfs->originds->dsid;
606 	zfs->originds->phys->ds_prev_snap_obj = zfs->snapds->dsid;
607 	zfs->originds->phys->ds_snapnames_zapobj = snapmapid;
608 
609 	zfs->snapds->phys->ds_next_snap_obj = zfs->originds->dsid;
610 	assert(zfs->snapds->phys->ds_num_children > 0);
611 	zfs->snapds->phys->ds_num_children++;
612 
613 	zap_write(zfs, zfs->cloneszap);
614 
615 	/* XXX-MJ dirs and datasets are leaked */
616 }
617 
618 void
dsl_dir_dataset_write(zfs_opt_t * zfs,zfs_objset_t * os,zfs_dsl_dir_t * dir)619 dsl_dir_dataset_write(zfs_opt_t *zfs, zfs_objset_t *os, zfs_dsl_dir_t *dir)
620 {
621 	dir->headds->os = os;
622 	objset_write(zfs, os);
623 }
624 
625 bool
dsl_dir_has_dataset(zfs_dsl_dir_t * dir)626 dsl_dir_has_dataset(zfs_dsl_dir_t *dir)
627 {
628 	return (dir->headds != NULL);
629 }
630 
631 bool
dsl_dir_dataset_has_objset(zfs_dsl_dir_t * dir)632 dsl_dir_dataset_has_objset(zfs_dsl_dir_t *dir)
633 {
634 	return (dsl_dir_has_dataset(dir) && dir->headds->os != NULL);
635 }
636 
637 static zfs_dsl_dataset_t *
dsl_dataset_alloc(zfs_opt_t * zfs,zfs_dsl_dir_t * dir)638 dsl_dataset_alloc(zfs_opt_t *zfs, zfs_dsl_dir_t *dir)
639 {
640 	zfs_dsl_dataset_t *ds;
641 	dnode_phys_t *dnode;
642 	uint64_t deadlistid;
643 
644 	ds = ecalloc(1, sizeof(*ds));
645 
646 	dnode = objset_dnode_bonus_alloc(zfs->mos, DMU_OT_DSL_DATASET,
647 	    DMU_OT_DSL_DATASET, sizeof(dsl_dataset_phys_t), &ds->dsid);
648 	ds->phys = (dsl_dataset_phys_t *)DN_BONUS(dnode);
649 
650 	dnode = objset_dnode_bonus_alloc(zfs->mos, DMU_OT_DEADLIST,
651 	    DMU_OT_DEADLIST_HDR, sizeof(dsl_deadlist_phys_t), &deadlistid);
652 	zap_write(zfs, zap_alloc(zfs->mos, dnode));
653 
654 	ds->phys->ds_dir_obj = dir->dirid;
655 	ds->phys->ds_deadlist_obj = deadlistid;
656 	ds->phys->ds_creation_txg = TXG - 1;
657 	if (ds != zfs->snapds)
658 		ds->phys->ds_prev_snap_txg = TXG - 1;
659 	ds->phys->ds_guid = randomguid();
660 	ds->dir = dir;
661 
662 	return (ds);
663 }
664