/*
* Open-context function to add one entry to the new mapping. The new
* entry will be remembered and written from syncing context.
*/
static void
spa_condense_indirect_commit_entry(spa_t *spa,
vdev_indirect_mapping_entry_phys_t *vimep, uint32_t count)
{
spa_condensing_indirect_t *sci = spa->spa_condensing_indirect;
ASSERT3U(count, <, DVA_GET_ASIZE(&vimep->vimep_dst));
dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
dmu_tx_hold_space(tx, sizeof (*vimep) + sizeof (count));
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
int txgoff = dmu_tx_get_txg(tx) & TXG_MASK;
/*
* If we are the first entry committed this txg, kick off the sync
* task to write to the MOS on our behalf.
*/
if (list_is_empty(&sci->sci_new_mapping_entries[txgoff])) {
dsl_sync_task_nowait(dmu_tx_pool(tx),
spa_condense_indirect_commit_sync, sci,
0, ZFS_SPACE_CHECK_NONE, tx);
}
vdev_indirect_mapping_entry_t *vime =
kmem_alloc(sizeof (*vime), KM_SLEEP);
vime->vime_mapping = *vimep;
vime->vime_obsolete_count = count;
list_insert_tail(&sci->sci_new_mapping_entries[txgoff], vime);
dmu_tx_commit(tx);
}
/*
* Ensure the zap is flushed then inform the VFS of the capacity change.
*/
static int
zvol_update_volsize(uint64_t volsize, objset_t *os)
{
dmu_tx_t *tx;
int error;
uint64_t txg;
ASSERT(MUTEX_HELD(&zvol_state_lock));
tx = dmu_tx_create(os);
dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
dmu_tx_mark_netfree(tx);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
return (SET_ERROR(error));
}
txg = dmu_tx_get_txg(tx);
error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
&volsize, tx);
dmu_tx_commit(tx);
txg_wait_synced(dmu_objset_pool(os), txg);
if (error == 0)
error = dmu_free_long_range(os,
ZVOL_OBJ, volsize, DMU_OBJECT_END);
return (error);
}
int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
sa_handle_t *sa_hdl, dmu_tx_t *tx,
struct lu_attr *la, uint64_t parent)
{
sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
uint64_t gen;
uint64_t crtime[2];
timestruc_t now;
int cnt;
int rc;
LASSERT(sa_hdl);
gen = dmu_tx_get_txg(tx);
gethrestime(&now);
ZFS_TIME_ENCODE(&now, crtime);
osa->atime[0] = la->la_atime;
osa->ctime[0] = la->la_ctime;
osa->mtime[0] = la->la_mtime;
osa->mode = la->la_mode;
osa->uid = la->la_uid;
osa->gid = la->la_gid;
osa->rdev = la->la_rdev;
osa->nlink = la->la_nlink;
osa->flags = attrs_fs2zfs(la->la_flags);
osa->size = la->la_size;
/*
* we need to create all SA below upon object create.
*
* XXX The attribute order matters since the accounting callback relies
* on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
* look up the UID/GID attributes. Moreover, the callback does not seem
* to support the spill block.
* We define attributes in the same order as SA_*_OFFSET in order to
* work around the problem. See ORI-610.
*/
cnt = 0;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
return rc;
}
/*
* This sync task completes (finishes) a condense, deleting the old
* mapping and replacing it with the new one.
*/
static void
spa_condense_indirect_complete_sync(void *arg, dmu_tx_t *tx)
{
spa_condensing_indirect_t *sci = arg;
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
spa_condensing_indirect_phys_t *scip =
&spa->spa_condensing_indirect_phys;
vdev_t *vd = vdev_lookup_top(spa, scip->scip_vdev);
vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
objset_t *mos = spa->spa_meta_objset;
vdev_indirect_mapping_t *old_mapping = vd->vdev_indirect_mapping;
uint64_t old_count = vdev_indirect_mapping_num_entries(old_mapping);
uint64_t new_count =
vdev_indirect_mapping_num_entries(sci->sci_new_mapping);
ASSERT(dmu_tx_is_syncing(tx));
ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
ASSERT3P(sci, ==, spa->spa_condensing_indirect);
for (int i = 0; i < TXG_SIZE; i++) {
ASSERT(list_is_empty(&sci->sci_new_mapping_entries[i]));
}
ASSERT(vic->vic_mapping_object != 0);
ASSERT3U(vd->vdev_id, ==, scip->scip_vdev);
ASSERT(scip->scip_next_mapping_object != 0);
ASSERT(scip->scip_prev_obsolete_sm_object != 0);
/*
* Reset vdev_indirect_mapping to refer to the new object.
*/
rw_enter(&vd->vdev_indirect_rwlock, RW_WRITER);
vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
vd->vdev_indirect_mapping = sci->sci_new_mapping;
rw_exit(&vd->vdev_indirect_rwlock);
sci->sci_new_mapping = NULL;
vdev_indirect_mapping_free(mos, vic->vic_mapping_object, tx);
vic->vic_mapping_object = scip->scip_next_mapping_object;
scip->scip_next_mapping_object = 0;
space_map_free_obj(mos, scip->scip_prev_obsolete_sm_object, tx);
spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
scip->scip_prev_obsolete_sm_object = 0;
scip->scip_vdev = 0;
VERIFY0(zap_remove(mos, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_CONDENSING_INDIRECT, tx));
spa_condensing_indirect_destroy(spa->spa_condensing_indirect);
spa->spa_condensing_indirect = NULL;
zfs_dbgmsg("finished condense of vdev %llu in txg %llu: "
"new mapping object %llu has %llu entries "
"(was %llu entries)",
vd->vdev_id, dmu_tx_get_txg(tx), vic->vic_mapping_object,
new_count, old_count);
vdev_config_dirty(spa->spa_root_vdev);
}
/*
* In one tx, free all log blocks and clear the log header.
* If keep_first is set, then we're replaying a log with no content.
* We want to keep the first block, however, so that the first
* synchronous transaction doesn't require a txg_wait_synced()
* in zil_create(). We don't need to txg_wait_synced() here either
* when keep_first is set, because both zil_create() and zil_destroy()
* will wait for any in-progress destroys to complete.
*/
void
zil_destroy(zilog_t *zilog, boolean_t keep_first)
{
const zil_header_t *zh = zilog->zl_header;
lwb_t *lwb;
dmu_tx_t *tx;
uint64_t txg;
/*
* Wait for any previous destroy to complete.
*/
txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
if (BP_IS_HOLE(&zh->zh_log))
return;
tx = dmu_tx_create(zilog->zl_os);
(void) dmu_tx_assign(tx, TXG_WAIT);
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
txg = dmu_tx_get_txg(tx);
mutex_enter(&zilog->zl_lock);
/*
* It is possible for the ZIL to get the previously mounted zilog
* structure of the same dataset if quickly remounted and the dbuf
* eviction has not completed. In this case we can see a non
* empty lwb list and keep_first will be set. We fix this by
* clearing the keep_first. This will be slower but it's very rare.
*/
if (!list_is_empty(&zilog->zl_lwb_list) && keep_first)
keep_first = B_FALSE;
ASSERT3U(zilog->zl_destroy_txg, <, txg);
zilog->zl_destroy_txg = txg;
zilog->zl_keep_first = keep_first;
if (!list_is_empty(&zilog->zl_lwb_list)) {
ASSERT(zh->zh_claim_txg == 0);
ASSERT(!keep_first);
while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
list_remove(&zilog->zl_lwb_list, lwb);
if (lwb->lwb_buf != NULL)
zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
zio_free_blk(zilog->zl_spa, &lwb->lwb_blk, txg);
kmem_cache_free(zil_lwb_cache, lwb);
}
} else {
if (!keep_first) {
(void) zil_parse(zilog, zil_free_log_block,
zil_free_log_record, tx, zh->zh_claim_txg);
}
}
mutex_exit(&zilog->zl_lock);
dmu_tx_commit(tx);
}
int
dsl_sync_task_group_wait(dsl_sync_task_group_t *dstg)
{
dmu_tx_t *tx;
uint64_t txg;
dsl_sync_task_t *dst;
top:
tx = dmu_tx_create_dd(dstg->dstg_pool->dp_mos_dir);
VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT));
txg = dmu_tx_get_txg(tx);
/* Do a preliminary error check. */
dstg->dstg_err = 0;
#ifdef ZFS_DEBUG
/*
* Only check half the time, otherwise, the sync-context
* check will almost never fail.
*/
if (spa_get_random(2) == 0)
goto skip;
#endif
rw_enter(&dstg->dstg_pool->dp_config_rwlock, RW_READER);
for (dst = list_head(&dstg->dstg_tasks); dst;
dst = list_next(&dstg->dstg_tasks, dst)) {
dst->dst_err =
dst->dst_checkfunc(dst->dst_arg1, dst->dst_arg2, tx);
if (dst->dst_err)
dstg->dstg_err = dst->dst_err;
}
rw_exit(&dstg->dstg_pool->dp_config_rwlock);
if (dstg->dstg_err) {
dmu_tx_commit(tx);
return (dstg->dstg_err);
}
skip:
/*
* We don't generally have many sync tasks, so pay the price of
* add_tail to get the tasks executed in the right order.
*/
VERIFY(0 == txg_list_add_tail(&dstg->dstg_pool->dp_sync_tasks,
dstg, txg));
dmu_tx_commit(tx);
txg_wait_synced(dstg->dstg_pool, txg);
if (dstg->dstg_err == EAGAIN) {
txg_wait_synced(dstg->dstg_pool, txg + TXG_DEFER_SIZE);
goto top;
}
return (dstg->dstg_err);
}
/* Takes care of physical writing and limiting # of concurrent ZIOs. */
static int
vdev_initialize_write(vdev_t *vd, uint64_t start, uint64_t size, abd_t *data)
{
spa_t *spa = vd->vdev_spa;
/* Limit inflight initializing I/Os */
mutex_enter(&vd->vdev_initialize_io_lock);
while (vd->vdev_initialize_inflight >= zfs_initialize_limit) {
cv_wait(&vd->vdev_initialize_io_cv,
&vd->vdev_initialize_io_lock);
}
vd->vdev_initialize_inflight++;
mutex_exit(&vd->vdev_initialize_io_lock);
dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
uint64_t txg = dmu_tx_get_txg(tx);
spa_config_enter(spa, SCL_STATE_ALL, vd, RW_READER);
mutex_enter(&vd->vdev_initialize_lock);
if (vd->vdev_initialize_offset[txg & TXG_MASK] == 0) {
uint64_t *guid = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
*guid = vd->vdev_guid;
/* This is the first write of this txg. */
dsl_sync_task_nowait(spa_get_dsl(spa),
vdev_initialize_zap_update_sync, guid, 2,
ZFS_SPACE_CHECK_RESERVED, tx);
}
/*
* We know the vdev struct will still be around since all
* consumers of vdev_free must stop the initialization first.
*/
if (vdev_initialize_should_stop(vd)) {
mutex_enter(&vd->vdev_initialize_io_lock);
ASSERT3U(vd->vdev_initialize_inflight, >, 0);
vd->vdev_initialize_inflight--;
mutex_exit(&vd->vdev_initialize_io_lock);
spa_config_exit(vd->vdev_spa, SCL_STATE_ALL, vd);
mutex_exit(&vd->vdev_initialize_lock);
dmu_tx_commit(tx);
return (SET_ERROR(EINTR));
}
mutex_exit(&vd->vdev_initialize_lock);
vd->vdev_initialize_offset[txg & TXG_MASK] = start + size;
zio_nowait(zio_write_phys(spa->spa_txg_zio[txg & TXG_MASK], vd, start,
size, data, ZIO_CHECKSUM_OFF, vdev_initialize_cb, NULL,
ZIO_PRIORITY_INITIALIZING, ZIO_FLAG_CANFAIL, B_FALSE));
/* vdev_initialize_cb releases SCL_STATE_ALL */
dmu_tx_commit(tx);
return (0);
}
void
dsl_sync_task_group_nowait(dsl_sync_task_group_t *dstg, dmu_tx_t *tx)
{
uint64_t txg;
dstg->dstg_nowaiter = B_TRUE;
txg = dmu_tx_get_txg(tx);
VERIFY(0 == txg_list_add(&dstg->dstg_pool->dp_sync_tasks, dstg, txg));
}
/*
* Called from open context to perform a callback in syncing context. Waits
* for the operation to complete.
*
* The checkfunc will be called from open context as a preliminary check
* which can quickly fail. If it succeeds, it will be called again from
* syncing context. The checkfunc should generally be designed to work
* properly in either context, but if necessary it can check
* dmu_tx_is_syncing(tx).
*
* The synctask infrastructure enforces proper locking strategy with respect
* to the dp_config_rwlock -- the lock will always be held when the callbacks
* are called. It will be held for read during the open-context (preliminary)
* call to the checkfunc, and then held for write from syncing context during
* the calls to the check and sync funcs.
*
* A dataset or pool name can be passed as the first argument. Typically,
* the check func will hold, check the return value of the hold, and then
* release the dataset. The sync func will VERIFYO(hold()) the dataset.
* This is safe because no changes can be made between the check and sync funcs,
* and the sync func will only be called if the check func successfully opened
* the dataset.
*/
int
dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
dsl_syncfunc_t *syncfunc, void *arg,
int blocks_modified, zfs_space_check_t space_check)
{
spa_t *spa;
dmu_tx_t *tx;
int err;
dsl_sync_task_t dst = { { { NULL } } };
dsl_pool_t *dp;
err = spa_open(pool, &spa, FTAG);
if (err != 0)
return (err);
dp = spa_get_dsl(spa);
top:
tx = dmu_tx_create_dd(dp->dp_mos_dir);
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
dst.dst_pool = dp;
dst.dst_txg = dmu_tx_get_txg(tx);
dst.dst_space = blocks_modified << DST_AVG_BLKSHIFT;
dst.dst_space_check = space_check;
dst.dst_checkfunc = checkfunc != NULL ? checkfunc : dsl_null_checkfunc;
dst.dst_syncfunc = syncfunc;
dst.dst_arg = arg;
dst.dst_error = 0;
dst.dst_nowaiter = B_FALSE;
dsl_pool_config_enter(dp, FTAG);
err = dst.dst_checkfunc(arg, tx);
dsl_pool_config_exit(dp, FTAG);
if (err != 0) {
dmu_tx_commit(tx);
spa_close(spa, FTAG);
return (err);
}
VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, &dst, dst.dst_txg));
dmu_tx_commit(tx);
txg_wait_synced(dp, dst.dst_txg);
if (dst.dst_error == EAGAIN) {
txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE);
goto top;
}
spa_close(spa, FTAG);
return (dst.dst_error);
}
int
zil_claim(char *osname, void *txarg)
{
dmu_tx_t *tx = txarg;
uint64_t first_txg = dmu_tx_get_txg(tx);
zilog_t *zilog;
zil_header_t *zh;
objset_t *os;
int error;
error = dmu_objset_open(osname, DMU_OST_ANY, DS_MODE_USER, &os);
if (error) {
cmn_err(CE_WARN, "can't open objset for %s", osname);
return (0);
}
zilog = dmu_objset_zil(os);
zh = zil_header_in_syncing_context(zilog);
/*
* Record here whether the zil has any records to replay.
* If the header block pointer is null or the block points
* to the stubby then we know there are no valid log records.
* We use the header to store this state as the the zilog gets
* freed later in dmu_objset_close().
* The flags (and the rest of the header fields) are cleared in
* zil_sync() as a result of a zil_destroy(), after replaying the log.
*
* Note, the intent log can be empty but still need the
* stubby to be claimed.
*/
if (!zil_empty(zilog))
zh->zh_flags |= ZIL_REPLAY_NEEDED;
/*
* Claim all log blocks if we haven't already done so, and remember
* the highest claimed sequence number. This ensures that if we can
* read only part of the log now (e.g. due to a missing device),
* but we can read the entire log later, we will not try to replay
* or destroy beyond the last block we successfully claimed.
*/
ASSERT3U(zh->zh_claim_txg, <=, first_txg);
if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) {
zh->zh_claim_txg = first_txg;
zh->zh_claim_seq = zil_parse(zilog, zil_claim_log_block,
zil_claim_log_record, tx, first_txg);
dsl_dataset_dirty(dmu_objset_ds(os), tx);
}
ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1));
dmu_objset_close(os);
return (0);
}
/*
* This sync task appends entries to the new mapping object.
*/
static void
spa_condense_indirect_commit_sync(void *arg, dmu_tx_t *tx)
{
spa_condensing_indirect_t *sci = arg;
uint64_t txg = dmu_tx_get_txg(tx);
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
ASSERT(dmu_tx_is_syncing(tx));
ASSERT3P(sci, ==, spa->spa_condensing_indirect);
vdev_indirect_mapping_add_entries(sci->sci_new_mapping,
&sci->sci_new_mapping_entries[txg & TXG_MASK], tx);
ASSERT(list_is_empty(&sci->sci_new_mapping_entries[txg & TXG_MASK]));
}
/*
* Create a new DMU object to hold a zfs znode.
*
* IN: dzp - parent directory for new znode
* vap - file attributes for new znode
* tx - dmu transaction id for zap operations
* cr - credentials of caller
* flag - flags:
* IS_ROOT_NODE - new object will be root
* IS_XATTR - new object is an attribute
* bonuslen - length of bonus buffer
* setaclp - File/Dir initial ACL
* fuidp - Tracks fuid allocation.
*
* OUT: zpp - allocated znode
*
*/
void
zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
{
uint64_t crtime[2], atime[2], mtime[2], ctime[2];
uint64_t mode, size, links, parent, pflags;
uint64_t dzp_pflags = 0;
uint64_t rdev = 0;
zfs_sb_t *zsb = ZTOZSB(dzp);
dmu_buf_t *db;
timestruc_t now;
uint64_t gen, obj;
int err;
int bonuslen;
sa_handle_t *sa_hdl;
dmu_object_type_t obj_type;
sa_bulk_attr_t *sa_attrs;
int cnt = 0;
zfs_acl_locator_cb_t locate = { 0 };
if (zsb->z_replay) {
obj = vap->va_nodeid;
now = vap->va_ctime; /* see zfs_replay_create() */
gen = vap->va_nblocks; /* ditto */
} else {
obj = 0;
gethrestime(&now);
gen = dmu_tx_get_txg(tx);
}
obj_type = zsb->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
bonuslen = (obj_type == DMU_OT_SA) ?
DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
/*
* Create a new DMU object.
*/
/*
* There's currently no mechanism for pre-reading the blocks that will
* be needed to allocate a new object, so we accept the small chance
* that there will be an i/o error and we will fail one of the
* assertions below.
*/
if (S_ISDIR(vap->va_mode)) {
if (zsb->z_replay) {
err = zap_create_claim_norm(zsb->z_os, obj,
zsb->z_norm, DMU_OT_DIRECTORY_CONTENTS,
obj_type, bonuslen, tx);
ASSERT3U(err, ==, 0);
} else {
void
dsl_sync_task_group_nowait(dsl_sync_task_group_t *dstg, dmu_tx_t *tx)
{
uint64_t txg;
dstg->dstg_nowaiter = B_TRUE;
txg = dmu_tx_get_txg(tx);
/*
* We don't generally have many sync tasks, so pay the price of
* add_tail to get the tasks executed in the right order.
*/
VERIFY(0 == txg_list_add_tail(&dstg->dstg_pool->dp_sync_tasks,
dstg, txg));
}
int
dsl_sync_task_group_wait(dsl_sync_task_group_t *dstg)
{
dmu_tx_t *tx;
uint64_t txg;
dsl_sync_task_t *dst;
top:
tx = dmu_tx_create_dd(dstg->dstg_pool->dp_mos_dir);
VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT));
txg = dmu_tx_get_txg(tx);
/* Do a preliminary error check. */
dstg->dstg_err = 0;
rw_enter(&dstg->dstg_pool->dp_config_rwlock, RW_READER);
for (dst = list_head(&dstg->dstg_tasks); dst;
dst = list_next(&dstg->dstg_tasks, dst)) {
#ifdef ZFS_DEBUG
/*
* Only check half the time, otherwise, the sync-context
* check will almost never fail.
*/
if (spa_get_random(2) == 0)
continue;
#endif
dst->dst_err =
dst->dst_checkfunc(dst->dst_arg1, dst->dst_arg2, tx);
if (dst->dst_err)
dstg->dstg_err = dst->dst_err;
}
rw_exit(&dstg->dstg_pool->dp_config_rwlock);
if (dstg->dstg_err) {
dmu_tx_commit(tx);
return (dstg->dstg_err);
}
VERIFY(0 == txg_list_add(&dstg->dstg_pool->dp_sync_tasks, dstg, txg));
dmu_tx_commit(tx);
txg_wait_synced(dstg->dstg_pool, txg);
if (dstg->dstg_err == EAGAIN)
goto top;
return (dstg->dstg_err);
}
/*
* Sync task to begin the condensing process.
*/
void
spa_condense_indirect_start_sync(vdev_t *vd, dmu_tx_t *tx)
{
spa_t *spa = vd->vdev_spa;
spa_condensing_indirect_phys_t *scip =
&spa->spa_condensing_indirect_phys;
ASSERT0(scip->scip_next_mapping_object);
ASSERT0(scip->scip_prev_obsolete_sm_object);
ASSERT0(scip->scip_vdev);
ASSERT(dmu_tx_is_syncing(tx));
ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
ASSERT(spa_feature_is_active(spa, SPA_FEATURE_OBSOLETE_COUNTS));
ASSERT(vdev_indirect_mapping_num_entries(vd->vdev_indirect_mapping));
uint64_t obsolete_sm_obj = vdev_obsolete_sm_object(vd);
ASSERT(obsolete_sm_obj != 0);
scip->scip_vdev = vd->vdev_id;
scip->scip_next_mapping_object =
vdev_indirect_mapping_alloc(spa->spa_meta_objset, tx);
scip->scip_prev_obsolete_sm_object = obsolete_sm_obj;
/*
* We don't need to allocate a new space map object, since
* vdev_indirect_sync_obsolete will allocate one when needed.
*/
space_map_close(vd->vdev_obsolete_sm);
vd->vdev_obsolete_sm = NULL;
VERIFY0(zap_remove(spa->spa_meta_objset, vd->vdev_top_zap,
VDEV_TOP_ZAP_INDIRECT_OBSOLETE_SM, tx));
VERIFY0(zap_add(spa->spa_dsl_pool->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_CONDENSING_INDIRECT, sizeof (uint64_t),
sizeof (*scip) / sizeof (uint64_t), scip, tx));
ASSERT3P(spa->spa_condensing_indirect, ==, NULL);
spa->spa_condensing_indirect = spa_condensing_indirect_create(spa);
zfs_dbgmsg("starting condense of vdev %llu in txg %llu: "
"posm=%llu nm=%llu",
vd->vdev_id, dmu_tx_get_txg(tx),
(u_longlong_t)scip->scip_prev_obsolete_sm_object,
(u_longlong_t)scip->scip_next_mapping_object);
zthr_wakeup(spa->spa_condense_zthr);
}
static void
zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg)
{
/*
* If we previously claimed it, we need to free it.
*/
if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE) {
lr_write_t *lr = (lr_write_t *)lrc;
blkptr_t *bp = &lr->lr_blkptr;
if (bp->blk_birth >= claim_txg &&
!zil_dva_tree_add(&zilog->zl_dva_tree, BP_IDENTITY(bp))) {
(void) arc_free(NULL, zilog->zl_spa,
dmu_tx_get_txg(tx), bp, NULL, NULL, ARC_WAIT);
}
}
}
uint64_t
zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx)
{
uint64_t seq;
ASSERT(itx->itx_lr.lrc_seq == 0);
mutex_enter(&zilog->zl_lock);
list_insert_tail(&zilog->zl_itx_list, itx);
zilog->zl_itx_list_sz += itx->itx_sod;
itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx);
itx->itx_lr.lrc_seq = seq = ++zilog->zl_itx_seq;
mutex_exit(&zilog->zl_lock);
return (seq);
}
void
dsl_sync_task_nowait(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg,
int blocks_modified, dmu_tx_t *tx)
{
dsl_sync_task_t *dst = kmem_zalloc(sizeof (*dst), KM_SLEEP);
dst->dst_pool = dp;
dst->dst_txg = dmu_tx_get_txg(tx);
dst->dst_space = blocks_modified << DST_AVG_BLKSHIFT;
dst->dst_checkfunc = dsl_null_checkfunc;
dst->dst_syncfunc = syncfunc;
dst->dst_arg = arg;
dst->dst_error = 0;
dst->dst_nowaiter = B_TRUE;
VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, dst, dst->dst_txg));
}
static void
vdev_initialize_zap_update_sync(void *arg, dmu_tx_t *tx)
{
/*
* We pass in the guid instead of the vdev_t since the vdev may
* have been freed prior to the sync task being processed. This
* happens when a vdev is detached as we call spa_config_vdev_exit(),
* stop the initializing thread, schedule the sync task, and free
* the vdev. Later when the scheduled sync task is invoked, it would
* find that the vdev has been freed.
*/
uint64_t guid = *(uint64_t *)arg;
uint64_t txg = dmu_tx_get_txg(tx);
kmem_free(arg, sizeof (uint64_t));
vdev_t *vd = spa_lookup_by_guid(tx->tx_pool->dp_spa, guid, B_FALSE);
if (vd == NULL || vd->vdev_top->vdev_removing || !vdev_is_concrete(vd))
return;
uint64_t last_offset = vd->vdev_initialize_offset[txg & TXG_MASK];
vd->vdev_initialize_offset[txg & TXG_MASK] = 0;
VERIFY(vd->vdev_leaf_zap != 0);
objset_t *mos = vd->vdev_spa->spa_meta_objset;
if (last_offset > 0) {
vd->vdev_initialize_last_offset = last_offset;
VERIFY0(zap_update(mos, vd->vdev_leaf_zap,
VDEV_LEAF_ZAP_INITIALIZE_LAST_OFFSET,
sizeof (last_offset), 1, &last_offset, tx));
}
if (vd->vdev_initialize_action_time > 0) {
uint64_t val = (uint64_t)vd->vdev_initialize_action_time;
VERIFY0(zap_update(mos, vd->vdev_leaf_zap,
VDEV_LEAF_ZAP_INITIALIZE_ACTION_TIME, sizeof (val),
1, &val, tx));
}
uint64_t initialize_state = vd->vdev_initialize_state;
VERIFY0(zap_update(mos, vd->vdev_leaf_zap,
VDEV_LEAF_ZAP_INITIALIZE_STATE, sizeof (initialize_state), 1,
&initialize_state, tx));
}
static void
free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx)
{
dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset;
uint64_t bytesfreed = 0;
int i;
dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num);
for (i = 0; i < num; i++, bp++) {
uint64_t lsize, lvl;
dmu_object_type_t type;
if (BP_IS_HOLE(bp))
continue;
bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE);
ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys));
/*
* Save some useful information on the holes being
* punched, including logical size, type, and indirection
* level. Retaining birth time enables detection of when
* holes are punched for reducing the number of free
* records transmitted during a zfs send.
*/
lsize = BP_GET_LSIZE(bp);
type = BP_GET_TYPE(bp);
lvl = BP_GET_LEVEL(bp);
bzero(bp, sizeof (blkptr_t));
if (spa_feature_is_active(dn->dn_objset->os_spa,
SPA_FEATURE_HOLE_BIRTH)) {
BP_SET_LSIZE(bp, lsize);
BP_SET_TYPE(bp, type);
BP_SET_LEVEL(bp, lvl);
BP_SET_BIRTH(bp, dmu_tx_get_txg(tx), 0);
}
}
dnode_diduse_space(dn, -bytesfreed);
}
/*
* Create a new DMU object to hold a zfs znode.
*
* IN: dzp - parent directory for new znode
* vap - file attributes for new znode
* tx - dmu transaction id for zap operations
* cr - credentials of caller
* flag - flags:
* IS_ROOT_NODE - new object will be root
* IS_XATTR - new object is an attribute
* IS_REPLAY - intent log replay
* bonuslen - length of bonus buffer
* setaclp - File/Dir initial ACL
* fuidp - Tracks fuid allocation.
*
* OUT: zpp - allocated znode
*
*/
void
zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
uint_t flag, znode_t **zpp, int bonuslen, zfs_acl_t *setaclp,
zfs_fuid_info_t **fuidp)
{
dmu_buf_t *db;
znode_phys_t *pzp;
zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
timestruc_t now;
uint64_t gen, obj;
int err;
ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
if (zfsvfs->z_assign >= TXG_INITIAL) { /* ZIL replay */
obj = vap->va_nodeid;
flag |= IS_REPLAY;
now = vap->va_ctime; /* see zfs_replay_create() */
gen = vap->va_nblocks; /* ditto */
} else {
obj = 0;
gethrestime(&now);
gen = dmu_tx_get_txg(tx);
}
/*
* Create a new DMU object.
*/
/*
* There's currently no mechanism for pre-reading the blocks that will
* be to needed allocate a new object, so we accept the small chance
* that there will be an i/o error and we will fail one of the
* assertions below.
*/
if (vap->va_type == VDIR) {
if (flag & IS_REPLAY) {
err = zap_create_claim_norm(zfsvfs->z_os, obj,
zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
DMU_OT_ZNODE, sizeof (znode_phys_t) + bonuslen, tx);
ASSERT3U(err, ==, 0);
} else {
/*
* Close an intent log.
*/
void
zil_close(zilog_t *zilog)
{
/*
* If the log isn't already committed, mark the objset dirty
* (so zil_sync() will be called) and wait for that txg to sync.
*/
if (!zil_is_committed(zilog)) {
uint64_t txg;
dmu_tx_t *tx = dmu_tx_create(zilog->zl_os);
(void) dmu_tx_assign(tx, TXG_WAIT);
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
txg = dmu_tx_get_txg(tx);
dmu_tx_commit(tx);
txg_wait_synced(zilog->zl_dmu_pool, txg);
}
taskq_destroy(zilog->zl_clean_taskq);
zilog->zl_clean_taskq = NULL;
zilog->zl_get_data = NULL;
zil_itx_clean(zilog);
ASSERT(list_head(&zilog->zl_itx_list) == NULL);
}
/*
* Create a new DMU object to hold a zfs znode.
*
* IN: dzp - parent directory for new znode
* vap - file attributes for new znode
* tx - dmu transaction id for zap operations
* cr - credentials of caller
* flag - flags:
* IS_ROOT_NODE - new object will be root
* IS_XATTR - new object is an attribute
* bonuslen - length of bonus buffer
* setaclp - File/Dir initial ACL
* fuidp - Tracks fuid allocation.
*
* OUT: zpp - allocated znode
*
*/
void
zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
{
uint64_t crtime[2], atime[2], mtime[2], ctime[2];
uint64_t mode, size, links, parent, pflags;
uint64_t dzp_pflags = 0;
uint64_t rdev = 0;
zfs_sb_t *zsb = ZTOZSB(dzp);
dmu_buf_t *db;
timestruc_t now;
uint64_t gen, obj;
int bonuslen;
sa_handle_t *sa_hdl;
dmu_object_type_t obj_type;
sa_bulk_attr_t *sa_attrs;
int cnt = 0;
zfs_acl_locator_cb_t locate = { 0 };
if (zsb->z_replay) {
obj = vap->va_nodeid;
now = vap->va_ctime; /* see zfs_replay_create() */
gen = vap->va_nblocks; /* ditto */
} else {
obj = 0;
gethrestime(&now);
gen = dmu_tx_get_txg(tx);
}
obj_type = zsb->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
bonuslen = (obj_type == DMU_OT_SA) ?
DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
/*
* Create a new DMU object.
*/
/*
* There's currently no mechanism for pre-reading the blocks that will
* be needed to allocate a new object, so we accept the small chance
* that there will be an i/o error and we will fail one of the
* assertions below.
*/
if (S_ISDIR(vap->va_mode)) {
if (zsb->z_replay) {
VERIFY0(zap_create_claim_norm(zsb->z_os, obj,
zsb->z_norm, DMU_OT_DIRECTORY_CONTENTS,
obj_type, bonuslen, tx));
} else {
obj = zap_create_norm(zsb->z_os,
zsb->z_norm, DMU_OT_DIRECTORY_CONTENTS,
obj_type, bonuslen, tx);
}
} else {
if (zsb->z_replay) {
VERIFY0(dmu_object_claim(zsb->z_os, obj,
DMU_OT_PLAIN_FILE_CONTENTS, 0,
obj_type, bonuslen, tx));
} else {
obj = dmu_object_alloc(zsb->z_os,
DMU_OT_PLAIN_FILE_CONTENTS, 0,
obj_type, bonuslen, tx);
}
}
ZFS_OBJ_HOLD_ENTER(zsb, obj);
VERIFY(0 == sa_buf_hold(zsb->z_os, obj, NULL, &db));
/*
* If this is the root, fix up the half-initialized parent pointer
* to reference the just-allocated physical data area.
*/
if (flag & IS_ROOT_NODE) {
dzp->z_id = obj;
} else {
dzp_pflags = dzp->z_pflags;
}
/*
* If parent is an xattr, so am I.
*/
if (dzp_pflags & ZFS_XATTR) {
flag |= IS_XATTR;
}
if (zsb->z_use_fuids)
pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
else
pflags = 0;
if (S_ISDIR(vap->va_mode)) {
size = 2; /* contents ("." and "..") */
links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
} else {
size = links = 0;
}
if (S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))
rdev = vap->va_rdev;
parent = dzp->z_id;
mode = acl_ids->z_mode;
if (flag & IS_XATTR)
pflags |= ZFS_XATTR;
/*
* No execs denied will be deterimed when zfs_mode_compute() is called.
*/
pflags |= acl_ids->z_aclp->z_hints &
(ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
ZFS_TIME_ENCODE(&now, crtime);
ZFS_TIME_ENCODE(&now, ctime);
if (vap->va_mask & ATTR_ATIME) {
ZFS_TIME_ENCODE(&vap->va_atime, atime);
} else {
ZFS_TIME_ENCODE(&now, atime);
}
if (vap->va_mask & ATTR_MTIME) {
ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
} else {
ZFS_TIME_ENCODE(&now, mtime);
}
/* Now add in all of the "SA" attributes */
VERIFY(0 == sa_handle_get_from_db(zsb->z_os, db, NULL, SA_HDL_SHARED,
&sa_hdl));
/*
* Setup the array of attributes to be replaced/set on the new file
*
* order for DMU_OT_ZNODE is critical since it needs to be constructed
* in the old znode_phys_t format. Don't change this ordering
*/
sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
if (obj_type == DMU_OT_ZNODE) {
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zsb),
NULL, &atime, 16);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zsb),
NULL, &mtime, 16);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zsb),
NULL, &ctime, 16);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zsb),
NULL, &crtime, 16);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zsb),
NULL, &gen, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zsb),
NULL, &mode, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zsb),
NULL, &size, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zsb),
NULL, &parent, 8);
} else {
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zsb),
NULL, &mode, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zsb),
NULL, &size, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zsb),
NULL, &gen, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zsb),
NULL, &acl_ids->z_fuid, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zsb),
NULL, &acl_ids->z_fgid, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zsb),
NULL, &parent, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zsb),
NULL, &pflags, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zsb),
NULL, &atime, 16);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zsb),
NULL, &mtime, 16);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zsb),
NULL, &ctime, 16);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zsb),
NULL, &crtime, 16);
}
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zsb), NULL, &links, 8);
if (obj_type == DMU_OT_ZNODE) {
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zsb), NULL,
&empty_xattr, 8);
}
if (obj_type == DMU_OT_ZNODE ||
(S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))) {
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zsb),
NULL, &rdev, 8);
}
if (obj_type == DMU_OT_ZNODE) {
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zsb),
NULL, &pflags, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zsb), NULL,
&acl_ids->z_fuid, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zsb), NULL,
&acl_ids->z_fgid, 8);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zsb), NULL, pad,
sizeof (uint64_t) * 4);
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zsb), NULL,
&acl_phys, sizeof (zfs_acl_phys_t));
} else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zsb), NULL,
&acl_ids->z_aclp->z_acl_count, 8);
locate.cb_aclp = acl_ids->z_aclp;
SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zsb),
zfs_acl_data_locator, &locate,
acl_ids->z_aclp->z_acl_bytes);
mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
acl_ids->z_fuid, acl_ids->z_fgid);
}
VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
if (!(flag & IS_ROOT_NODE)) {
*zpp = zfs_znode_alloc(zsb, db, 0, obj_type, obj, sa_hdl,
ZTOI(dzp));
VERIFY(*zpp != NULL);
VERIFY(dzp != NULL);
} else {
/*
* If we are creating the root node, the "parent" we
* passed in is the znode for the root.
*/
*zpp = dzp;
(*zpp)->z_sa_hdl = sa_hdl;
}
(*zpp)->z_pflags = pflags;
(*zpp)->z_mode = mode;
if (obj_type == DMU_OT_ZNODE ||
acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
}
kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
ZFS_OBJ_HOLD_EXIT(zsb, obj);
}
/*
* Called in syncing context to free committed log blocks and update log header.
*/
void
zil_sync(zilog_t *zilog, dmu_tx_t *tx)
{
zil_header_t *zh = zil_header_in_syncing_context(zilog);
uint64_t txg = dmu_tx_get_txg(tx);
spa_t *spa = zilog->zl_spa;
lwb_t *lwb;
/*
* We don't zero out zl_destroy_txg, so make sure we don't try
* to destroy it twice.
*/
if (spa_sync_pass(spa) != 1)
return;
mutex_enter(&zilog->zl_lock);
ASSERT(zilog->zl_stop_sync == 0);
zh->zh_replay_seq = zilog->zl_replayed_seq[txg & TXG_MASK];
if (zilog->zl_destroy_txg == txg) {
blkptr_t blk = zh->zh_log;
ASSERT(list_head(&zilog->zl_lwb_list) == NULL);
bzero(zh, sizeof (zil_header_t));
bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq));
if (zilog->zl_keep_first) {
/*
* If this block was part of log chain that couldn't
* be claimed because a device was missing during
* zil_claim(), but that device later returns,
* then this block could erroneously appear valid.
* To guard against this, assign a new GUID to the new
* log chain so it doesn't matter what blk points to.
*/
zil_init_log_chain(zilog, &blk);
zh->zh_log = blk;
}
}
while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
zh->zh_log = lwb->lwb_blk;
if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg)
break;
list_remove(&zilog->zl_lwb_list, lwb);
zio_free_blk(spa, &lwb->lwb_blk, txg);
kmem_cache_free(zil_lwb_cache, lwb);
/*
* If we don't have anything left in the lwb list then
* we've had an allocation failure and we need to zero
* out the zil_header blkptr so that we don't end
* up freeing the same block twice.
*/
if (list_head(&zilog->zl_lwb_list) == NULL)
BP_ZERO(&zh->zh_log);
}
mutex_exit(&zilog->zl_lock);
}
int __osd_attr_init(const struct lu_env *env, udmu_objset_t *uos,
uint64_t oid, dmu_tx_t *tx, struct lu_attr *la)
{
sa_bulk_attr_t *bulk;
sa_handle_t *sa_hdl;
struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
uint64_t gen;
uint64_t parent;
uint64_t crtime[2];
timestruc_t now;
int cnt;
int rc;
gethrestime(&now);
gen = dmu_tx_get_txg(tx);
ZFS_TIME_ENCODE(&now, crtime);
/* XXX: this should be real id of parent for ZPL access, but we have no
* such info in OSD, probably it can be part of dt_object_format */
parent = 0;
osa->atime[0] = la->la_atime;
osa->ctime[0] = la->la_ctime;
osa->mtime[0] = la->la_mtime;
osa->mode = la->la_mode;
osa->uid = la->la_uid;
osa->gid = la->la_gid;
osa->rdev = la->la_rdev;
osa->nlink = la->la_nlink;
osa->flags = la->la_flags;
osa->size = la->la_size;
/* Now add in all of the "SA" attributes */
rc = -sa_handle_get(uos->os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
if (rc)
return rc;
OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
if (bulk == NULL) {
rc = -ENOMEM;
goto out;
}
/*
* we need to create all SA below upon object create.
*
* XXX The attribute order matters since the accounting callback relies
* on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
* look up the UID/GID attributes. Moreover, the callback does not seem
* to support the spill block.
* We define attributes in the same order as SA_*_OFFSET in order to
* work around the problem. See ORI-610.
*/
cnt = 0;
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(uos), NULL, &gen, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(uos), NULL, &parent, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(uos), NULL, crtime, 16);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL, &osa->rdev, 8);
rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
out:
sa_handle_destroy(sa_hdl);
return rc;
}
int
zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t len,
char *addr)
{
dmu_object_info_t doi;
ssize_t nbytes;
itx_t *itx;
lr_write_t *lr;
objset_t *os;
dmu_buf_t *db;
uint64_t txg;
uint64_t boff;
int error;
uint32_t blocksize;
/* handle common case */
if (len <= zvol_immediate_write_sz) {
itx = zvol_immediate_itx(off, len, addr);
(void) zil_itx_assign(zv->zv_zilog, itx, tx);
return (0);
}
txg = dmu_tx_get_txg(tx);
os = zv->zv_objset;
/*
* We need to dmu_sync() each block in the range.
* For this we need the blocksize.
*/
error = dmu_object_info(os, ZVOL_OBJ, &doi);
if (error)
return (error);
blocksize = doi.doi_data_block_size;
/*
* We need to immediate write or dmu_sync() each block in the range.
*/
while (len) {
nbytes = MIN(len, blocksize - P2PHASE(off, blocksize));
if (nbytes <= zvol_immediate_write_sz) {
itx = zvol_immediate_itx(off, nbytes, addr);
} else {
boff = P2ALIGN_TYPED(off, blocksize, uint64_t);
itx = zil_itx_create(TX_WRITE, sizeof (*lr));
lr = (lr_write_t *)&itx->itx_lr;
lr->lr_foid = ZVOL_OBJ;
lr->lr_offset = off;
lr->lr_length = nbytes;
lr->lr_blkoff = off - boff;
BP_ZERO(&lr->lr_blkptr);
/* XXX - we should do these IOs in parallel */
VERIFY(0 == dmu_buf_hold(os, ZVOL_OBJ, boff,
FTAG, &db));
ASSERT(boff == db->db_offset);
error = dmu_sync(NULL, db, &lr->lr_blkptr,
txg, NULL, NULL);
dmu_buf_rele(db, FTAG);
if (error) {
kmem_free(itx, offsetof(itx_t, itx_lr));
return (error);
}
itx->itx_wr_state = WR_COPIED;
}
(void) zil_itx_assign(zv->zv_zilog, itx, tx);
len -= nbytes;
off += nbytes;
}
return (0);
}
/* ARGSUSED */
static void
zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg)
{
zio_free_blk(zilog->zl_spa, bp, dmu_tx_get_txg(tx));
}
/*
* Create an on-disk intent log.
*/
static void
zil_create(zilog_t *zilog)
{
const zil_header_t *zh = zilog->zl_header;
lwb_t *lwb;
uint64_t txg = 0;
dmu_tx_t *tx = NULL;
blkptr_t blk;
int error = 0;
/*
* Wait for any previous destroy to complete.
*/
txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
ASSERT(zh->zh_claim_txg == 0);
ASSERT(zh->zh_replay_seq == 0);
blk = zh->zh_log;
/*
* If we don't already have an initial log block or we have one
* but it's the wrong endianness then allocate one.
*/
if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) {
tx = dmu_tx_create(zilog->zl_os);
(void) dmu_tx_assign(tx, TXG_WAIT);
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
txg = dmu_tx_get_txg(tx);
if (!BP_IS_HOLE(&blk)) {
zio_free_blk(zilog->zl_spa, &blk, txg);
BP_ZERO(&blk);
}
error = zio_alloc_blk(zilog->zl_spa, ZIL_MIN_BLKSZ, &blk,
NULL, txg);
if (error == 0)
zil_init_log_chain(zilog, &blk);
}
/*
* Allocate a log write buffer (lwb) for the first log block.
*/
if (error == 0) {
lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP);
lwb->lwb_zilog = zilog;
lwb->lwb_blk = blk;
lwb->lwb_nused = 0;
lwb->lwb_sz = BP_GET_LSIZE(&lwb->lwb_blk);
lwb->lwb_buf = zio_buf_alloc(lwb->lwb_sz);
lwb->lwb_max_txg = txg;
lwb->lwb_zio = NULL;
mutex_enter(&zilog->zl_lock);
list_insert_tail(&zilog->zl_lwb_list, lwb);
mutex_exit(&zilog->zl_lock);
}
/*
* If we just allocated the first log block, commit our transaction
* and wait for zil_sync() to stuff the block poiner into zh_log.
* (zh is part of the MOS, so we cannot modify it in open context.)
*/
if (tx != NULL) {
dmu_tx_commit(tx);
txg_wait_synced(zilog->zl_dmu_pool, txg);
}
ASSERT(bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0);
}
