*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
*/
#include <sys/zfs_context.h>
uint64_t td_objset;
blkptr_t *td_rootbp;
uint64_t td_min_txg;
+ zbookmark_t *td_resume;
int td_flags;
prefetch_data_t *td_pfd;
blkptr_cb_t *td_func;
zil_free(zilog);
}
+typedef enum resume_skip {
+ RESUME_SKIP_ALL,
+ RESUME_SKIP_NONE,
+ RESUME_SKIP_CHILDREN
+} resume_skip_t;
+
+/*
+ * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and
+ * the block indicated by zb does not need to be visited at all. Returns
+ * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the
+ * resume point. This indicates that this block should be visited but not its
+ * children (since they must have been visited in a previous traversal).
+ * Otherwise returns RESUME_SKIP_NONE.
+ */
+static resume_skip_t
+resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp,
+ const zbookmark_t *zb)
+{
+ if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) {
+ /*
+ * If we already visited this bp & everything below,
+ * don't bother doing it again.
+ */
+ if (zbookmark_is_before(dnp, zb, td->td_resume))
+ return (RESUME_SKIP_ALL);
+
+ /*
+ * If we found the block we're trying to resume from, zero
+ * the bookmark out to indicate that we have resumed.
+ */
+ ASSERT3U(zb->zb_object, <=, td->td_resume->zb_object);
+ if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) {
+ bzero(td->td_resume, sizeof (*zb));
+ if (td->td_flags & TRAVERSE_POST)
+ return (RESUME_SKIP_CHILDREN);
+ }
+ }
+ return (RESUME_SKIP_NONE);
+}
+
+static void
+traverse_pause(traverse_data_t *td, const zbookmark_t *zb)
+{
+ ASSERT(td->td_resume != NULL);
+ ASSERT3U(zb->zb_level, ==, 0);
+ bcopy(zb, td->td_resume, sizeof (*td->td_resume));
+}
+
static int
traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb)
arc_buf_t *buf = NULL;
prefetch_data_t *pd = td->td_pfd;
boolean_t hard = td->td_flags & TRAVERSE_HARD;
+ boolean_t pause = B_FALSE;
- if (bp->blk_birth == 0) {
+ switch (resume_skip_check(td, dnp, zb)) {
+ case RESUME_SKIP_ALL:
+ return (0);
+ case RESUME_SKIP_CHILDREN:
+ goto post;
+ case RESUME_SKIP_NONE:
+ break;
+ default:
+ ASSERT(0);
+ }
+
+ if (BP_IS_HOLE(bp)) {
err = td->td_func(td->td_spa, NULL, NULL, pbuf, zb, dnp,
td->td_arg);
return (err);
td->td_arg);
if (err == TRAVERSE_VISIT_NO_CHILDREN)
return (0);
- if (err)
- return (err);
+ if (err == ERESTART)
+ pause = B_TRUE; /* handle pausing at a common point */
+ if (err != 0)
+ goto post;
}
if (BP_GET_LEVEL(bp) > 0) {
if (buf)
(void) arc_buf_remove_ref(buf, &buf);
+post:
if (err == 0 && lasterr == 0 && (td->td_flags & TRAVERSE_POST)) {
err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp,
td->td_arg);
+ if (err == ERESTART)
+ pause = B_TRUE;
+ }
+
+ if (pause && td->td_resume != NULL) {
+ ASSERT3U(err, ==, ERESTART);
+ ASSERT(!hard);
+ traverse_pause(td, zb);
}
return (err != 0 ? err : lasterr);
* in syncing context).
*/
static int
-traverse_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *rootbp,
- uint64_t txg_start, int flags, blkptr_cb_t func, void *arg)
+traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
+ uint64_t txg_start, zbookmark_t *resume, int flags,
+ blkptr_cb_t func, void *arg)
{
- traverse_data_t td;
- prefetch_data_t pd = { 0 };
- zbookmark_t czb;
+ traverse_data_t *td;
+ prefetch_data_t *pd;
+ zbookmark_t *czb;
int err;
- td.td_spa = spa;
- td.td_objset = ds ? ds->ds_object : 0;
- td.td_rootbp = rootbp;
- td.td_min_txg = txg_start;
- td.td_func = func;
- td.td_arg = arg;
- td.td_pfd = &pd;
- td.td_flags = flags;
+ ASSERT(ds == NULL || objset == ds->ds_object);
+ ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST));
+
+ td = kmem_alloc(sizeof(traverse_data_t), KM_PUSHPAGE);
+ pd = kmem_zalloc(sizeof(prefetch_data_t), KM_PUSHPAGE);
+ czb = kmem_alloc(sizeof(zbookmark_t), KM_PUSHPAGE);
- pd.pd_blks_max = zfs_pd_blks_max;
- pd.pd_flags = flags;
- mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL);
- cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL);
+ td->td_spa = spa;
+ td->td_objset = objset;
+ td->td_rootbp = rootbp;
+ td->td_min_txg = txg_start;
+ td->td_resume = resume;
+ td->td_func = func;
+ td->td_arg = arg;
+ td->td_pfd = pd;
+ td->td_flags = flags;
+
+ pd->pd_blks_max = zfs_pd_blks_max;
+ pd->pd_flags = flags;
+ mutex_init(&pd->pd_mtx, NULL, MUTEX_DEFAULT, NULL);
+ cv_init(&pd->pd_cv, NULL, CV_DEFAULT, NULL);
/* See comment on ZIL traversal in dsl_scan_visitds. */
if (ds != NULL && !dsl_dataset_is_snapshot(ds)) {
if (err)
return (err);
- traverse_zil(&td, &os->os_zil_header);
+ traverse_zil(td, &os->os_zil_header);
}
if (!(flags & TRAVERSE_PREFETCH) ||
0 == taskq_dispatch(system_taskq, traverse_prefetch_thread,
- &td, TQ_NOQUEUE))
- pd.pd_exited = B_TRUE;
+ td, TQ_NOQUEUE))
+ pd->pd_exited = B_TRUE;
- SET_BOOKMARK(&czb, td.td_objset,
+ SET_BOOKMARK(czb, td->td_objset,
ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
- err = traverse_visitbp(&td, NULL, NULL, rootbp, &czb);
+ err = traverse_visitbp(td, NULL, NULL, rootbp, czb);
- mutex_enter(&pd.pd_mtx);
- pd.pd_cancel = B_TRUE;
- cv_broadcast(&pd.pd_cv);
- while (!pd.pd_exited)
- cv_wait(&pd.pd_cv, &pd.pd_mtx);
- mutex_exit(&pd.pd_mtx);
+ mutex_enter(&pd->pd_mtx);
+ pd->pd_cancel = B_TRUE;
+ cv_broadcast(&pd->pd_cv);
+ while (!pd->pd_exited)
+ cv_wait(&pd->pd_cv, &pd->pd_mtx);
+ mutex_exit(&pd->pd_mtx);
- mutex_destroy(&pd.pd_mtx);
- cv_destroy(&pd.pd_cv);
+ mutex_destroy(&pd->pd_mtx);
+ cv_destroy(&pd->pd_cv);
+
+ kmem_free(czb, sizeof(zbookmark_t));
+ kmem_free(pd, sizeof(struct prefetch_data));
+ kmem_free(td, sizeof(struct traverse_data));
return (err);
}
traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags,
blkptr_cb_t func, void *arg)
{
- return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds,
- &ds->ds_phys->ds_bp, txg_start, flags, func, arg));
+ return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object,
+ &ds->ds_phys->ds_bp, txg_start, NULL, flags, func, arg));
+}
+
+int
+traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr,
+ uint64_t txg_start, zbookmark_t *resume, int flags,
+ blkptr_cb_t func, void *arg)
+{
+ return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET,
+ blkptr, txg_start, resume, flags, func, arg));
}
/*
boolean_t hard = (flags & TRAVERSE_HARD);
/* visit the MOS */
- err = traverse_impl(spa, NULL, spa_get_rootblkptr(spa),
- txg_start, flags, func, arg);
+ err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa),
+ txg_start, NULL, flags, func, arg);
if (err)
return (err);
err = 0;
return (err != 0 ? err : lasterr);
}
+
+#if defined(_KERNEL) && defined(HAVE_SPL)
+EXPORT_SYMBOL(traverse_dataset);
+EXPORT_SYMBOL(traverse_pool);
+
+module_param(zfs_pd_blks_max, int, 0644);
+MODULE_PARM_DESC(zfs_pd_blks_max, "Max number of blocks to prefetch");
+#endif