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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 */
24
25#include <sys/zfs_context.h>
26#include <sys/dmu_objset.h>
27#include <sys/dmu_traverse.h>
28#include <sys/dsl_dataset.h>
29#include <sys/dsl_dir.h>
30#include <sys/dsl_pool.h>
31#include <sys/dnode.h>
32#include <sys/spa.h>
33#include <sys/zio.h>
34#include <sys/dmu_impl.h>
35#include <sys/sa.h>
36#include <sys/sa_impl.h>
37#include <sys/callb.h>
38
39int zfs_pd_blks_max = 100;
40
41typedef struct prefetch_data {
42	kmutex_t pd_mtx;
43	kcondvar_t pd_cv;
44	int pd_blks_max;
45	int pd_blks_fetched;
46	int pd_flags;
47	boolean_t pd_cancel;
48	boolean_t pd_exited;
49} prefetch_data_t;
50
51typedef struct traverse_data {
52	spa_t *td_spa;
53	uint64_t td_objset;
54	blkptr_t *td_rootbp;
55	uint64_t td_min_txg;
56	int td_flags;
57	prefetch_data_t *td_pfd;
58	blkptr_cb_t *td_func;
59	void *td_arg;
60} traverse_data_t;
61
62static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
63    arc_buf_t *buf, uint64_t objset, uint64_t object);
64
65static int
66traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
67{
68	traverse_data_t *td = arg;
69	zbookmark_t zb;
70
71	if (bp->blk_birth == 0)
72		return (0);
73
74	if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
75		return (0);
76
77	SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
78	    bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
79
80	(void) td->td_func(td->td_spa, zilog, bp, NULL, &zb, NULL, td->td_arg);
81
82	return (0);
83}
84
85static int
86traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
87{
88	traverse_data_t *td = arg;
89
90	if (lrc->lrc_txtype == TX_WRITE) {
91		lr_write_t *lr = (lr_write_t *)lrc;
92		blkptr_t *bp = &lr->lr_blkptr;
93		zbookmark_t zb;
94
95		if (bp->blk_birth == 0)
96			return (0);
97
98		if (claim_txg == 0 || bp->blk_birth < claim_txg)
99			return (0);
100
101		SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid,
102		    ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
103
104		(void) td->td_func(td->td_spa, zilog, bp, NULL, &zb, NULL,
105		    td->td_arg);
106	}
107	return (0);
108}
109
110static void
111traverse_zil(traverse_data_t *td, zil_header_t *zh)
112{
113	uint64_t claim_txg = zh->zh_claim_txg;
114	zilog_t *zilog;
115
116	/*
117	 * We only want to visit blocks that have been claimed but not yet
118	 * replayed; plus, in read-only mode, blocks that are already stable.
119	 */
120	if (claim_txg == 0 && spa_writeable(td->td_spa))
121		return;
122
123	zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);
124
125	(void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
126	    claim_txg);
127
128	zil_free(zilog);
129}
130
131static int
132traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
133    arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb)
134{
135	zbookmark_t czb;
136	int err = 0, lasterr = 0;
137	arc_buf_t *buf = NULL;
138	prefetch_data_t *pd = td->td_pfd;
139	boolean_t hard = td->td_flags & TRAVERSE_HARD;
140
141	if (bp->blk_birth == 0) {
142		err = td->td_func(td->td_spa, NULL, NULL, pbuf, zb, dnp,
143		    td->td_arg);
144		return (err);
145	}
146
147	if (bp->blk_birth <= td->td_min_txg)
148		return (0);
149
150	if (pd && !pd->pd_exited &&
151	    ((pd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
152	    BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) {
153		mutex_enter(&pd->pd_mtx);
154		ASSERT(pd->pd_blks_fetched >= 0);
155		while (pd->pd_blks_fetched == 0 && !pd->pd_exited)
156			cv_wait(&pd->pd_cv, &pd->pd_mtx);
157		pd->pd_blks_fetched--;
158		cv_broadcast(&pd->pd_cv);
159		mutex_exit(&pd->pd_mtx);
160	}
161
162	if (td->td_flags & TRAVERSE_PRE) {
163		err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp,
164		    td->td_arg);
165		if (err == TRAVERSE_VISIT_NO_CHILDREN)
166			return (0);
167		if (err)
168			return (err);
169	}
170
171	if (BP_GET_LEVEL(bp) > 0) {
172		uint32_t flags = ARC_WAIT;
173		int i;
174		blkptr_t *cbp;
175		int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
176
177		err = dsl_read(NULL, td->td_spa, bp, pbuf,
178		    arc_getbuf_func, &buf,
179		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
180		if (err)
181			return (err);
182
183		/* recursively visitbp() blocks below this */
184		cbp = buf->b_data;
185		for (i = 0; i < epb; i++, cbp++) {
186			SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
187			    zb->zb_level - 1,
188			    zb->zb_blkid * epb + i);
189			err = traverse_visitbp(td, dnp, buf, cbp, &czb);
190			if (err) {
191				if (!hard)
192					break;
193				lasterr = err;
194			}
195		}
196	} else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
197		uint32_t flags = ARC_WAIT;
198		int i;
199		int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
200
201		err = dsl_read(NULL, td->td_spa, bp, pbuf,
202		    arc_getbuf_func, &buf,
203		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
204		if (err)
205			return (err);
206
207		/* recursively visitbp() blocks below this */
208		dnp = buf->b_data;
209		for (i = 0; i < epb; i++, dnp++) {
210			err = traverse_dnode(td, dnp, buf, zb->zb_objset,
211			    zb->zb_blkid * epb + i);
212			if (err) {
213				if (!hard)
214					break;
215				lasterr = err;
216			}
217		}
218	} else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
219		uint32_t flags = ARC_WAIT;
220		objset_phys_t *osp;
221		dnode_phys_t *dnp;
222
223		err = dsl_read_nolock(NULL, td->td_spa, bp,
224		    arc_getbuf_func, &buf,
225		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
226		if (err)
227			return (err);
228
229		osp = buf->b_data;
230		dnp = &osp->os_meta_dnode;
231		err = traverse_dnode(td, dnp, buf, zb->zb_objset,
232		    DMU_META_DNODE_OBJECT);
233		if (err && hard) {
234			lasterr = err;
235			err = 0;
236		}
237		if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
238			dnp = &osp->os_userused_dnode;
239			err = traverse_dnode(td, dnp, buf, zb->zb_objset,
240			    DMU_USERUSED_OBJECT);
241		}
242		if (err && hard) {
243			lasterr = err;
244			err = 0;
245		}
246		if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
247			dnp = &osp->os_groupused_dnode;
248			err = traverse_dnode(td, dnp, buf, zb->zb_objset,
249			    DMU_GROUPUSED_OBJECT);
250		}
251	}
252
253	if (buf)
254		(void) arc_buf_remove_ref(buf, &buf);
255
256	if (err == 0 && lasterr == 0 && (td->td_flags & TRAVERSE_POST)) {
257		err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp,
258		    td->td_arg);
259	}
260
261	return (err != 0 ? err : lasterr);
262}
263
264static int
265traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
266    arc_buf_t *buf, uint64_t objset, uint64_t object)
267{
268	int j, err = 0, lasterr = 0;
269	zbookmark_t czb;
270	boolean_t hard = (td->td_flags & TRAVERSE_HARD);
271
272	for (j = 0; j < dnp->dn_nblkptr; j++) {
273		SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
274		err = traverse_visitbp(td, dnp, buf,
275		    (blkptr_t *)&dnp->dn_blkptr[j], &czb);
276		if (err) {
277			if (!hard)
278				break;
279			lasterr = err;
280		}
281	}
282
283	if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
284		SET_BOOKMARK(&czb, objset,
285		    object, 0, DMU_SPILL_BLKID);
286		err = traverse_visitbp(td, dnp, buf,
287		    (blkptr_t *)&dnp->dn_spill, &czb);
288		if (err) {
289			if (!hard)
290				return (err);
291			lasterr = err;
292		}
293	}
294	return (err != 0 ? err : lasterr);
295}
296
297/* ARGSUSED */
298static int
299traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
300    arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp,
301    void *arg)
302{
303	prefetch_data_t *pfd = arg;
304	uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
305
306	ASSERT(pfd->pd_blks_fetched >= 0);
307	if (pfd->pd_cancel)
308		return (EINTR);
309
310	if (bp == NULL || !((pfd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
311	    BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0) ||
312	    BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
313		return (0);
314
315	mutex_enter(&pfd->pd_mtx);
316	while (!pfd->pd_cancel && pfd->pd_blks_fetched >= pfd->pd_blks_max)
317		cv_wait(&pfd->pd_cv, &pfd->pd_mtx);
318	pfd->pd_blks_fetched++;
319	cv_broadcast(&pfd->pd_cv);
320	mutex_exit(&pfd->pd_mtx);
321
322	(void) dsl_read(NULL, spa, bp, pbuf, NULL, NULL,
323	    ZIO_PRIORITY_ASYNC_READ,
324	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
325	    &aflags, zb);
326
327	return (0);
328}
329
330static void
331traverse_prefetch_thread(void *arg)
332{
333	traverse_data_t *td_main = arg;
334	traverse_data_t td = *td_main;
335	zbookmark_t czb;
336
337	td.td_func = traverse_prefetcher;
338	td.td_arg = td_main->td_pfd;
339	td.td_pfd = NULL;
340
341	SET_BOOKMARK(&czb, td.td_objset,
342	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
343	(void) traverse_visitbp(&td, NULL, NULL, td.td_rootbp, &czb);
344
345	mutex_enter(&td_main->td_pfd->pd_mtx);
346	td_main->td_pfd->pd_exited = B_TRUE;
347	cv_broadcast(&td_main->td_pfd->pd_cv);
348	mutex_exit(&td_main->td_pfd->pd_mtx);
349}
350
351/*
352 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
353 * in syncing context).
354 */
355static int
356traverse_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *rootbp,
357    uint64_t txg_start, int flags, blkptr_cb_t func, void *arg)
358{
359	traverse_data_t td;
360	prefetch_data_t pd = { 0 };
361	zbookmark_t czb;
362	int err;
363
364	td.td_spa = spa;
365	td.td_objset = ds ? ds->ds_object : 0;
366	td.td_rootbp = rootbp;
367	td.td_min_txg = txg_start;
368	td.td_func = func;
369	td.td_arg = arg;
370	td.td_pfd = &pd;
371	td.td_flags = flags;
372
373	pd.pd_blks_max = zfs_pd_blks_max;
374	pd.pd_flags = flags;
375	mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL);
376	cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL);
377
378	/* See comment on ZIL traversal in dsl_scan_visitds. */
379	if (ds != NULL && !dsl_dataset_is_snapshot(ds)) {
380		objset_t *os;
381
382		err = dmu_objset_from_ds(ds, &os);
383		if (err)
384			return (err);
385
386		traverse_zil(&td, &os->os_zil_header);
387	}
388
389	if (!(flags & TRAVERSE_PREFETCH) ||
390	    0 == taskq_dispatch(system_taskq, traverse_prefetch_thread,
391	    &td, TQ_NOQUEUE))
392		pd.pd_exited = B_TRUE;
393
394	SET_BOOKMARK(&czb, td.td_objset,
395	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
396	err = traverse_visitbp(&td, NULL, NULL, rootbp, &czb);
397
398	mutex_enter(&pd.pd_mtx);
399	pd.pd_cancel = B_TRUE;
400	cv_broadcast(&pd.pd_cv);
401	while (!pd.pd_exited)
402		cv_wait(&pd.pd_cv, &pd.pd_mtx);
403	mutex_exit(&pd.pd_mtx);
404
405	mutex_destroy(&pd.pd_mtx);
406	cv_destroy(&pd.pd_cv);
407
408	return (err);
409}
410
411/*
412 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
413 * in syncing context).
414 */
415int
416traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags,
417    blkptr_cb_t func, void *arg)
418{
419	return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds,
420	    &ds->ds_phys->ds_bp, txg_start, flags, func, arg));
421}
422
423/*
424 * NB: pool must not be changing on-disk (eg, from zdb or sync context).
425 */
426int
427traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
428    blkptr_cb_t func, void *arg)
429{
430	int err, lasterr = 0;
431	uint64_t obj;
432	dsl_pool_t *dp = spa_get_dsl(spa);
433	objset_t *mos = dp->dp_meta_objset;
434	boolean_t hard = (flags & TRAVERSE_HARD);
435
436	/* visit the MOS */
437	err = traverse_impl(spa, NULL, spa_get_rootblkptr(spa),
438	    txg_start, flags, func, arg);
439	if (err)
440		return (err);
441
442	/* visit each dataset */
443	for (obj = 1; err == 0 || (err != ESRCH && hard);
444	    err = dmu_object_next(mos, &obj, FALSE, txg_start)) {
445		dmu_object_info_t doi;
446
447		err = dmu_object_info(mos, obj, &doi);
448		if (err) {
449			if (!hard)
450				return (err);
451			lasterr = err;
452			continue;
453		}
454
455		if (doi.doi_type == DMU_OT_DSL_DATASET) {
456			dsl_dataset_t *ds;
457			uint64_t txg = txg_start;
458
459			rw_enter(&dp->dp_config_rwlock, RW_READER);
460			err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
461			rw_exit(&dp->dp_config_rwlock);
462			if (err) {
463				if (!hard)
464					return (err);
465				lasterr = err;
466				continue;
467			}
468			if (ds->ds_phys->ds_prev_snap_txg > txg)
469				txg = ds->ds_phys->ds_prev_snap_txg;
470			err = traverse_dataset(ds, txg, flags, func, arg);
471			dsl_dataset_rele(ds, FTAG);
472			if (err) {
473				if (!hard)
474					return (err);
475				lasterr = err;
476			}
477		}
478	}
479	if (err == ESRCH)
480		err = 0;
481	return (err != 0 ? err : lasterr);
482}
483