xref: /titanic_41/usr/src/uts/common/fs/zfs/dmu_traverse.c (revision e9af4bc0b1cc30cea75d6ad4aa2fde97d985e9be)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/zfs_context.h>
27 #include <sys/dmu_objset.h>
28 #include <sys/dmu_traverse.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_pool.h>
32 #include <sys/dnode.h>
33 #include <sys/spa.h>
34 #include <sys/zio.h>
35 #include <sys/dmu_impl.h>
36 #include <sys/callb.h>
37 
38 struct prefetch_data {
39 	kmutex_t pd_mtx;
40 	kcondvar_t pd_cv;
41 	int pd_blks_max;
42 	int pd_blks_fetched;
43 	int pd_flags;
44 	boolean_t pd_cancel;
45 	boolean_t pd_exited;
46 };
47 
48 struct traverse_data {
49 	spa_t *td_spa;
50 	uint64_t td_objset;
51 	blkptr_t *td_rootbp;
52 	uint64_t td_min_txg;
53 	int td_flags;
54 	struct prefetch_data *td_pfd;
55 	blkptr_cb_t *td_func;
56 	void *td_arg;
57 };
58 
59 static int traverse_dnode(struct traverse_data *td, const dnode_phys_t *dnp,
60     arc_buf_t *buf, uint64_t objset, uint64_t object);
61 
62 /* ARGSUSED */
63 static int
64 traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
65 {
66 	struct traverse_data *td = arg;
67 	zbookmark_t zb;
68 
69 	if (bp->blk_birth == 0)
70 		return (0);
71 
72 	if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
73 		return (0);
74 
75 	SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
76 	    bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
77 
78 	(void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg);
79 
80 	return (0);
81 }
82 
83 /* ARGSUSED */
84 static int
85 traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
86 {
87 	struct traverse_data *td = arg;
88 
89 	if (lrc->lrc_txtype == TX_WRITE) {
90 		lr_write_t *lr = (lr_write_t *)lrc;
91 		blkptr_t *bp = &lr->lr_blkptr;
92 		zbookmark_t zb;
93 
94 		if (bp->blk_birth == 0)
95 			return (0);
96 
97 		if (claim_txg == 0 || bp->blk_birth < claim_txg)
98 			return (0);
99 
100 		SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid, ZB_ZIL_LEVEL,
101 		    lr->lr_offset / BP_GET_LSIZE(bp));
102 
103 		(void) td->td_func(td->td_spa, zilog, bp, &zb, NULL,
104 		    td->td_arg);
105 	}
106 	return (0);
107 }
108 
109 static void
110 traverse_zil(struct traverse_data *td, zil_header_t *zh)
111 {
112 	uint64_t claim_txg = zh->zh_claim_txg;
113 	zilog_t *zilog;
114 
115 	/*
116 	 * We only want to visit blocks that have been claimed but not yet
117 	 * replayed; plus, in read-only mode, blocks that are already stable.
118 	 */
119 	if (claim_txg == 0 && spa_writeable(td->td_spa))
120 		return;
121 
122 	zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);
123 
124 	(void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
125 	    claim_txg);
126 
127 	zil_free(zilog);
128 }
129 
130 static int
131 traverse_visitbp(struct traverse_data *td, const dnode_phys_t *dnp,
132     arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb)
133 {
134 	zbookmark_t czb;
135 	int err = 0;
136 	arc_buf_t *buf = NULL;
137 	struct prefetch_data *pd = td->td_pfd;
138 
139 	if (bp->blk_birth == 0) {
140 		err = td->td_func(td->td_spa, NULL, NULL, zb, dnp, td->td_arg);
141 		return (err);
142 	}
143 
144 	if (bp->blk_birth <= td->td_min_txg)
145 		return (0);
146 
147 	if (pd && !pd->pd_exited &&
148 	    ((pd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
149 	    BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) {
150 		mutex_enter(&pd->pd_mtx);
151 		ASSERT(pd->pd_blks_fetched >= 0);
152 		while (pd->pd_blks_fetched == 0 && !pd->pd_exited)
153 			cv_wait(&pd->pd_cv, &pd->pd_mtx);
154 		pd->pd_blks_fetched--;
155 		cv_broadcast(&pd->pd_cv);
156 		mutex_exit(&pd->pd_mtx);
157 	}
158 
159 	if (td->td_flags & TRAVERSE_PRE) {
160 		err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
161 		if (err)
162 			return (err);
163 	}
164 
165 	if (BP_GET_LEVEL(bp) > 0) {
166 		uint32_t flags = ARC_WAIT;
167 		int i;
168 		blkptr_t *cbp;
169 		int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
170 
171 		err = arc_read(NULL, td->td_spa, bp, pbuf,
172 		    arc_getbuf_func, &buf,
173 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
174 		if (err)
175 			return (err);
176 
177 		/* recursively visitbp() blocks below this */
178 		cbp = buf->b_data;
179 		for (i = 0; i < epb; i++, cbp++) {
180 			SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
181 			    zb->zb_level - 1,
182 			    zb->zb_blkid * epb + i);
183 			err = traverse_visitbp(td, dnp, buf, cbp, &czb);
184 			if (err)
185 				break;
186 		}
187 	} else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
188 		uint32_t flags = ARC_WAIT;
189 		int i;
190 		int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
191 
192 		err = arc_read(NULL, td->td_spa, bp, pbuf,
193 		    arc_getbuf_func, &buf,
194 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
195 		if (err)
196 			return (err);
197 
198 		/* recursively visitbp() blocks below this */
199 		dnp = buf->b_data;
200 		for (i = 0; i < epb && err == 0; i++, dnp++) {
201 			err = traverse_dnode(td, dnp, buf, zb->zb_objset,
202 			    zb->zb_blkid * epb + i);
203 			if (err)
204 				break;
205 		}
206 	} else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
207 		uint32_t flags = ARC_WAIT;
208 		objset_phys_t *osp;
209 		dnode_phys_t *dnp;
210 
211 		err = arc_read_nolock(NULL, td->td_spa, bp,
212 		    arc_getbuf_func, &buf,
213 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
214 		if (err)
215 			return (err);
216 
217 		osp = buf->b_data;
218 		traverse_zil(td, &osp->os_zil_header);
219 
220 		dnp = &osp->os_meta_dnode;
221 		err = traverse_dnode(td, dnp, buf, zb->zb_objset,
222 		    DMU_META_DNODE_OBJECT);
223 		if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
224 			dnp = &osp->os_userused_dnode;
225 			err = traverse_dnode(td, dnp, buf, zb->zb_objset,
226 			    DMU_USERUSED_OBJECT);
227 		}
228 		if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
229 			dnp = &osp->os_groupused_dnode;
230 			err = traverse_dnode(td, dnp, buf, zb->zb_objset,
231 			    DMU_GROUPUSED_OBJECT);
232 		}
233 	}
234 
235 	if (buf)
236 		(void) arc_buf_remove_ref(buf, &buf);
237 
238 	if (err == 0 && (td->td_flags & TRAVERSE_POST))
239 		err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
240 
241 	return (err);
242 }
243 
244 static int
245 traverse_dnode(struct traverse_data *td, const dnode_phys_t *dnp,
246     arc_buf_t *buf, uint64_t objset, uint64_t object)
247 {
248 	int j, err = 0;
249 	zbookmark_t czb;
250 
251 	for (j = 0; j < dnp->dn_nblkptr; j++) {
252 		SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
253 		err = traverse_visitbp(td, dnp, buf,
254 		    (blkptr_t *)&dnp->dn_blkptr[j], &czb);
255 		if (err)
256 			break;
257 	}
258 	return (err);
259 }
260 
261 /* ARGSUSED */
262 static int
263 traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
264     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
265 {
266 	struct prefetch_data *pfd = arg;
267 	uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
268 
269 	ASSERT(pfd->pd_blks_fetched >= 0);
270 	if (pfd->pd_cancel)
271 		return (EINTR);
272 
273 	if (bp == NULL || !((pfd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
274 	    BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0))
275 		return (0);
276 
277 	mutex_enter(&pfd->pd_mtx);
278 	while (!pfd->pd_cancel && pfd->pd_blks_fetched >= pfd->pd_blks_max)
279 		cv_wait(&pfd->pd_cv, &pfd->pd_mtx);
280 	pfd->pd_blks_fetched++;
281 	cv_broadcast(&pfd->pd_cv);
282 	mutex_exit(&pfd->pd_mtx);
283 
284 	(void) arc_read_nolock(NULL, spa, bp, NULL, NULL,
285 	    ZIO_PRIORITY_ASYNC_READ,
286 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
287 	    &aflags, zb);
288 
289 	return (0);
290 }
291 
292 static void
293 traverse_prefetch_thread(void *arg)
294 {
295 	struct traverse_data *td_main = arg;
296 	struct traverse_data td = *td_main;
297 	zbookmark_t czb;
298 
299 	td.td_func = traverse_prefetcher;
300 	td.td_arg = td_main->td_pfd;
301 	td.td_pfd = NULL;
302 
303 	SET_BOOKMARK(&czb, td.td_objset,
304 	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
305 	(void) traverse_visitbp(&td, NULL, NULL, td.td_rootbp, &czb);
306 
307 	mutex_enter(&td_main->td_pfd->pd_mtx);
308 	td_main->td_pfd->pd_exited = B_TRUE;
309 	cv_broadcast(&td_main->td_pfd->pd_cv);
310 	mutex_exit(&td_main->td_pfd->pd_mtx);
311 }
312 
313 /*
314  * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
315  * in syncing context).
316  */
317 static int
318 traverse_impl(spa_t *spa, uint64_t objset, blkptr_t *rootbp,
319     uint64_t txg_start, int flags, blkptr_cb_t func, void *arg)
320 {
321 	struct traverse_data td;
322 	struct prefetch_data pd = { 0 };
323 	zbookmark_t czb;
324 	int err;
325 
326 	td.td_spa = spa;
327 	td.td_objset = objset;
328 	td.td_rootbp = rootbp;
329 	td.td_min_txg = txg_start;
330 	td.td_func = func;
331 	td.td_arg = arg;
332 	td.td_pfd = &pd;
333 	td.td_flags = flags;
334 
335 	pd.pd_blks_max = 100;
336 	pd.pd_flags = flags;
337 	mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL);
338 	cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL);
339 
340 	if (!(flags & TRAVERSE_PREFETCH) ||
341 	    0 == taskq_dispatch(system_taskq, traverse_prefetch_thread,
342 	    &td, TQ_NOQUEUE))
343 		pd.pd_exited = B_TRUE;
344 
345 	SET_BOOKMARK(&czb, objset,
346 	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
347 	err = traverse_visitbp(&td, NULL, NULL, rootbp, &czb);
348 
349 	mutex_enter(&pd.pd_mtx);
350 	pd.pd_cancel = B_TRUE;
351 	cv_broadcast(&pd.pd_cv);
352 	while (!pd.pd_exited)
353 		cv_wait(&pd.pd_cv, &pd.pd_mtx);
354 	mutex_exit(&pd.pd_mtx);
355 
356 	mutex_destroy(&pd.pd_mtx);
357 	cv_destroy(&pd.pd_cv);
358 
359 	return (err);
360 }
361 
362 /*
363  * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
364  * in syncing context).
365  */
366 int
367 traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags,
368     blkptr_cb_t func, void *arg)
369 {
370 	return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds->ds_object,
371 	    &ds->ds_phys->ds_bp, txg_start, flags, func, arg));
372 }
373 
374 /*
375  * NB: pool must not be changing on-disk (eg, from zdb or sync context).
376  */
377 int
378 traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
379     blkptr_cb_t func, void *arg)
380 {
381 	int err;
382 	uint64_t obj;
383 	dsl_pool_t *dp = spa_get_dsl(spa);
384 	objset_t *mos = dp->dp_meta_objset;
385 
386 	/* visit the MOS */
387 	err = traverse_impl(spa, 0, spa_get_rootblkptr(spa),
388 	    txg_start, flags, func, arg);
389 	if (err)
390 		return (err);
391 
392 	/* visit each dataset */
393 	for (obj = 1; err == 0; err = dmu_object_next(mos, &obj, FALSE,
394 	    txg_start)) {
395 		dmu_object_info_t doi;
396 
397 		err = dmu_object_info(mos, obj, &doi);
398 		if (err)
399 			return (err);
400 
401 		if (doi.doi_type == DMU_OT_DSL_DATASET) {
402 			dsl_dataset_t *ds;
403 			uint64_t txg = txg_start;
404 
405 			rw_enter(&dp->dp_config_rwlock, RW_READER);
406 			err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
407 			rw_exit(&dp->dp_config_rwlock);
408 			if (err)
409 				return (err);
410 			if (ds->ds_phys->ds_prev_snap_txg > txg)
411 				txg = ds->ds_phys->ds_prev_snap_txg;
412 			err = traverse_dataset(ds, txg, flags, func, arg);
413 			dsl_dataset_rele(ds, FTAG);
414 			if (err)
415 				return (err);
416 		}
417 	}
418 	if (err == ESRCH)
419 		err = 0;
420 	return (err);
421 }
422