/*
* Move console messages from src to dst. The time of day isn't known
* early in boot, so fix up the message timestamps if necessary.
*/
static void
log_conswitch(log_t *src, log_t *dst)
{
mblk_t *mp;
mblk_t *hmp = NULL;
mblk_t *tmp = NULL;
log_ctl_t *hlc;
while ((mp = getq_noenab(src->log_q, 0)) != NULL) {
log_ctl_t *lc = (log_ctl_t *)mp->b_rptr;
lc->flags |= SL_LOGONLY;
/*
* The ttime is written with 0 in log_sensmsg() only when
* good gethrestime_sec() data is not available to store in
* the log_ctl_t in the early boot phase.
*/
if (lc->ttime == 0) {
/*
* Look ahead to first early boot message with time.
*/
if (hmp) {
tmp->b_next = mp;
tmp = mp;
} else
hmp = tmp = mp;
continue;
}
while (hmp) {
tmp = hmp->b_next;
hmp->b_next = NULL;
hlc = (log_ctl_t *)hmp->b_rptr;
/*
* Calculate hrestime for an early log message with
* an invalid time stamp. We know:
* - the lbolt of the invalid time stamp.
* - the hrestime and lbolt of the first valid
* time stamp.
*/
hlc->ttime = lc->ttime - (lc->ltime - hlc->ltime) / hz;
(void) putq(dst->log_q, hmp);
hmp = tmp;
}
(void) putq(dst->log_q, mp);
}
while (hmp) {
tmp = hmp->b_next;
hmp->b_next = NULL;
hlc = (log_ctl_t *)hmp->b_rptr;
hlc->ttime = gethrestime_sec() -
(ddi_get_lbolt() - hlc->ltime) / hz;
(void) putq(dst->log_q, hmp);
hmp = tmp;
}
dst->log_overflow = src->log_overflow;
src->log_flags = 0;
dst->log_flags = SL_CONSOLE;
log_consq = dst->log_q;
}
/* ARGSUSED */
int
socket_vop_setattr(struct vnode *vp, struct vattr *vap, int flags,
struct cred *cr, caller_context_t *ct)
{
struct sonode *so = VTOSO(vp);
/*
* If times were changed, and we have a STREAMS socket, then update
* the sonode.
*/
if (!SOCK_IS_NONSTR(so)) {
sotpi_info_t *sti = SOTOTPI(so);
mutex_enter(&so->so_lock);
if (vap->va_mask & AT_ATIME)
sti->sti_atime = vap->va_atime.tv_sec;
if (vap->va_mask & AT_MTIME) {
sti->sti_mtime = vap->va_mtime.tv_sec;
sti->sti_ctime = gethrestime_sec();
}
mutex_exit(&so->so_lock);
}
return (0);
}
uint64_t
dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
dmu_tx_t *tx)
{
objset_t *mos = dp->dp_meta_objset;
uint64_t ddobj;
dsl_dir_phys_t *dsphys;
dmu_buf_t *dbuf;
ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
if (pds) {
VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
name, sizeof (uint64_t), 1, &ddobj, tx));
} else {
/* it's the root dir */
VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
}
VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
dmu_buf_will_dirty(dbuf, tx);
dsphys = dbuf->db_data;
dsphys->dd_creation_time = gethrestime_sec();
if (pds)
dsphys->dd_parent_obj = pds->dd_object;
dsphys->dd_props_zapobj = zap_create(mos,
DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
dsphys->dd_child_dir_zapobj = zap_create(mos,
DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
dmu_buf_rele(dbuf, FTAG);
return (ddobj);
}
/*
* XDR loopback unix auth parameters.
* NOTE: this is an XDR_ENCODE only routine.
*/
bool_t
xdr_authloopback(XDR *xdrs)
{
uid_t uid;
gid_t gid;
int len;
caddr_t groups;
char *name = uts_nodename();
struct cred *cr;
time_t now;
if (xdrs->x_op != XDR_ENCODE)
return (FALSE);
cr = CRED();
uid = crgetuid(cr);
gid = crgetgid(cr);
len = crgetngroups(cr);
groups = (caddr_t)crgetgroups(cr);
now = gethrestime_sec();
if (xdr_uint32(xdrs, (uint32_t *)&now) &&
xdr_string(xdrs, &name, MAX_MACHINE_NAME) &&
xdr_uid_t(xdrs, &uid) &&
xdr_gid_t(xdrs, &gid) &&
xdr_array(xdrs, &groups, (uint_t *)&len, NGRPS_LOOPBACK,
sizeof (int), (xdrproc_t)xdr_int))
return (TRUE);
return (FALSE);
}
static void
dsl_dataset_user_hold_sync(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_hold_arg_t *dduha = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvlist_t *tmpholds;
nvpair_t *pair;
uint64_t now = gethrestime_sec();
if (dduha->dduha_minor != 0)
tmpholds = fnvlist_alloc();
else
tmpholds = NULL;
for (pair = nvlist_next_nvpair(dduha->dduha_chkholds, NULL);
pair != NULL;
pair = nvlist_next_nvpair(dduha->dduha_chkholds, pair)) {
dsl_dataset_t *ds;
VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
dsl_dataset_user_hold_sync_one_impl(tmpholds, ds,
fnvpair_value_string(pair), dduha->dduha_minor, now, tx);
dsl_dataset_rele(ds, FTAG);
}
dsl_onexit_hold_cleanup(dp->dp_spa, tmpholds, dduha->dduha_minor);
}
static int
nd_auth_marshall(XDR *xdrp)
{
int credsize;
int32_t *ptr;
int hostnamelen;
hostnamelen = (int)strlen(utsname.nodename);
credsize = 4 + 4 + roundup(hostnamelen, 4) + 4 + 4 + 4;
ptr = XDR_INLINE(xdrp, 4 + 4 + credsize + 4 + 4);
if (!ptr) {
cmn_err(CE_WARN, "\tnfs_dump: auth_marshall failed");
return (0);
}
/*
* We can do the fast path.
*/
IXDR_PUT_INT32(ptr, AUTH_UNIX); /* cred flavor */
IXDR_PUT_INT32(ptr, credsize); /* cred len */
IXDR_PUT_INT32(ptr, gethrestime_sec());
IXDR_PUT_INT32(ptr, hostnamelen);
bcopy(utsname.nodename, ptr, hostnamelen);
ptr += roundup(hostnamelen, 4) / 4;
IXDR_PUT_INT32(ptr, 0); /* uid */
IXDR_PUT_INT32(ptr, 0); /* gid */
IXDR_PUT_INT32(ptr, 0); /* gid list length (empty) */
IXDR_PUT_INT32(ptr, AUTH_NULL); /* verf flavor */
IXDR_PUT_INT32(ptr, 0); /* verf len */
return (1);
}
/*
* msgget system call.
*/
static int
msgget(key_t key, int msgflg)
{
kmsqid_t *qp;
kmutex_t *lock;
int id, error;
int ii;
proc_t *pp = curproc;
top:
if (error = ipc_get(msq_svc, key, msgflg, (kipc_perm_t **)&qp, &lock))
return (set_errno(error));
if (IPC_FREE(&qp->msg_perm)) {
mutex_exit(lock);
mutex_exit(&pp->p_lock);
list_create(&qp->msg_list, sizeof (struct msg),
offsetof(struct msg, msg_node));
qp->msg_qnum = 0;
qp->msg_lspid = qp->msg_lrpid = 0;
qp->msg_stime = qp->msg_rtime = 0;
qp->msg_ctime = gethrestime_sec();
qp->msg_ngt_cnt = 0;
qp->msg_neg_copy = 0;
for (ii = 0; ii <= MSG_MAX_QNUM; ii++) {
list_create(&qp->msg_wait_snd[ii],
sizeof (msgq_wakeup_t),
offsetof(msgq_wakeup_t, msgw_list));
list_create(&qp->msg_wait_snd_ngt[ii],
sizeof (msgq_wakeup_t),
offsetof(msgq_wakeup_t, msgw_list));
}
/*
* The proper initialization of msg_lowest_type is to the
* highest possible value. By doing this we guarantee that
* when the first send happens, the lowest type will be set
* properly.
*/
qp->msg_lowest_type = -1;
list_create(&qp->msg_cpy_block,
sizeof (msgq_wakeup_t),
offsetof(msgq_wakeup_t, msgw_list));
qp->msg_fnd_sndr = &msg_fnd_sndr[0];
qp->msg_fnd_rdr = &msg_fnd_rdr[0];
qp->msg_rcv_cnt = 0;
qp->msg_snd_cnt = 0;
if (error = ipc_commit_begin(msq_svc, key, msgflg,
(kipc_perm_t *)qp)) {
if (error == EAGAIN)
goto top;
return (set_errno(error));
}
qp->msg_qbytes = rctl_enforced_value(rc_process_msgmnb,
pp->p_rctls, pp);
qp->msg_qmax = rctl_enforced_value(rc_process_msgtql,
pp->p_rctls, pp);
lock = ipc_commit_end(msq_svc, &qp->msg_perm);
}
/*
* Print these messages by running:
* echo ::zfs_dbgmsg | mdb -k
*
* Monitor these messages by running:
* dtrace -q -n 'zfs-dbgmsg{printf("%s\n", stringof(arg0))}'
*/
void
zfs_dbgmsg(const char *fmt, ...)
{
int size;
va_list adx;
zfs_dbgmsg_t *zdm;
va_start(adx, fmt);
size = vsnprintf(NULL, 0, fmt, adx);
va_end(adx);
/*
* There is one byte of string in sizeof (zfs_dbgmsg_t), used
* for the terminating null.
*/
zdm = kmem_alloc(sizeof (zfs_dbgmsg_t) + size, KM_SLEEP);
zdm->zdm_timestamp = gethrestime_sec();
va_start(adx, fmt);
(void) vsnprintf(zdm->zdm_msg, size + 1, fmt, adx);
va_end(adx);
DTRACE_PROBE1(zfs__dbgmsg, char *, zdm->zdm_msg);
mutex_enter(&zfs_dbgmsgs_lock);
list_insert_tail(&zfs_dbgmsgs, zdm);
zfs_dbgmsg_size += sizeof (zfs_dbgmsg_t) + size;
while (zfs_dbgmsg_size > zfs_dbgmsg_maxsize) {
zdm = list_remove_head(&zfs_dbgmsgs);
size = sizeof (zfs_dbgmsg_t) + strlen(zdm->zdm_msg);
kmem_free(zdm, size);
zfs_dbgmsg_size -= size;
}
mutex_exit(&zfs_dbgmsgs_lock);
}
/*
* When the uberblock on-disk is updated by a spa_sync,
* creating a new "best" uberblock, update the one stored
* in the mmp thread state, used for mmp writes.
*/
void
mmp_update_uberblock(spa_t *spa, uberblock_t *ub)
{
mmp_thread_t *mmp = &spa->spa_mmp;
mutex_enter(&mmp->mmp_io_lock);
mmp->mmp_ub = *ub;
mmp->mmp_ub.ub_timestamp = gethrestime_sec();
mutex_exit(&mmp->mmp_io_lock);
}
static void
vdev_initialize_change_state(vdev_t *vd, vdev_initializing_state_t new_state)
{
ASSERT(MUTEX_HELD(&vd->vdev_initialize_lock));
spa_t *spa = vd->vdev_spa;
if (new_state == vd->vdev_initialize_state)
return;
/*
* Copy the vd's guid, this will be freed by the sync task.
*/
uint64_t *guid = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
*guid = vd->vdev_guid;
/*
* If we're suspending, then preserving the original start time.
*/
if (vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED) {
vd->vdev_initialize_action_time = gethrestime_sec();
}
vd->vdev_initialize_state = new_state;
dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
dsl_sync_task_nowait(spa_get_dsl(spa), vdev_initialize_zap_update_sync,
guid, 2, ZFS_SPACE_CHECK_RESERVED, tx);
switch (new_state) {
case VDEV_INITIALIZE_ACTIVE:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s activated", vd->vdev_path);
break;
case VDEV_INITIALIZE_SUSPENDED:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s suspended", vd->vdev_path);
break;
case VDEV_INITIALIZE_CANCELED:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s canceled", vd->vdev_path);
break;
case VDEV_INITIALIZE_COMPLETE:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s complete", vd->vdev_path);
break;
default:
panic("invalid state %llu", (unsigned long long)new_state);
}
dmu_tx_commit(tx);
}
void
rfs4_recall_deleg(rfs4_file_t *fp, bool_t trunc, rfs4_client_t *cp)
{
time_t elapsed1, elapsed2;
if (fp->rf_dinfo.rd_time_recalled != 0) {
elapsed1 = gethrestime_sec() - fp->rf_dinfo.rd_time_recalled;
elapsed2 = gethrestime_sec() - fp->rf_dinfo.rd_time_lastwrite;
/* First check to see if a revocation should occur */
if (elapsed1 > rfs4_lease_time &&
elapsed2 > rfs4_lease_time) {
rfs4_revoke_file(fp);
return;
}
/*
* Next check to see if a recall should be done again
* so quickly.
*/
if (elapsed1 <= ((rfs4_lease_time * 20) / 100))
return;
}
rfs4_recall_file(fp, rfs4_do_cb_recall, trunc, cp);
}
/*
* Choose a random vdev, label, and MMP block, and write over it
* with a copy of the last-synced uberblock, whose timestamp
* has been updated to reflect that the pool is in use.
*/
static void
mmp_write_uberblock(spa_t *spa)
{
int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
mmp_thread_t *mmp = &spa->spa_mmp;
uberblock_t *ub;
vdev_t *vd;
int label;
uint64_t offset;
spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
vd = mmp_random_leaf(spa->spa_root_vdev);
if (vd == NULL) {
spa_config_exit(spa, SCL_STATE, FTAG);
return;
}
mutex_enter(&mmp->mmp_io_lock);
if (mmp->mmp_zio_root == NULL)
mmp->mmp_zio_root = zio_root(spa, NULL, NULL,
flags | ZIO_FLAG_GODFATHER);
ub = &mmp->mmp_ub;
ub->ub_timestamp = gethrestime_sec();
ub->ub_mmp_magic = MMP_MAGIC;
ub->ub_mmp_delay = mmp->mmp_delay;
vd->vdev_mmp_pending = gethrtime();
zio_t *zio = zio_null(mmp->mmp_zio_root, spa, NULL, NULL, NULL, flags);
abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
mutex_exit(&mmp->mmp_io_lock);
offset = VDEV_UBERBLOCK_OFFSET(vd, VDEV_UBERBLOCK_COUNT(vd) -
MMP_BLOCKS_PER_LABEL + spa_get_random(MMP_BLOCKS_PER_LABEL));
label = spa_get_random(VDEV_LABELS);
vdev_label_write(zio, vd, label, ub_abd, offset,
VDEV_UBERBLOCK_SIZE(vd), mmp_write_done, mmp,
flags | ZIO_FLAG_DONT_PROPAGATE);
spa_mmp_history_add(ub->ub_txg, ub->ub_timestamp, ub->ub_mmp_delay, vd,
label);
zio_nowait(zio);
}
static void
sys_log(const char *msg)
{
static time_t tstamp = 0;
time_t now;
/*
* msg is shown (at most) once per minute
*/
now = gethrestime_sec();
if ((tstamp + 60) < now) {
tstamp = now;
cmn_err(CE_WARN, msg);
}
}
static void
dsl_dataset_user_hold_sync(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_hold_arg_t *dduha = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvpair_t *pair;
uint64_t now = gethrestime_sec();
for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) {
dsl_dataset_t *ds;
VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
dsl_dataset_user_hold_sync_one(ds, fnvpair_value_string(pair),
dduha->dduha_minor, now, tx);
dsl_dataset_rele(ds, FTAG);
}
}
/*
* Allocate zeroed memory if tmpfs_maxkmem has not been exceeded
* or the 'musthave' flag is set. 'musthave' allocations should
* always be subordinate to normal allocations so that tmpfs_maxkmem
* can't be exceeded by more than a few KB. Example: when creating
* a new directory, the tmpnode is a normal allocation; if that
* succeeds, the dirents for "." and ".." are 'musthave' allocations.
*/
void *
tmp_memalloc(size_t size, int musthave)
{
static time_t last_warning;
time_t now;
if (atomic_add_long_nv(&tmp_kmemspace, size) < tmpfs_maxkmem ||
musthave)
return (kmem_zalloc(size, KM_SLEEP));
atomic_add_long(&tmp_kmemspace, -size);
now = gethrestime_sec();
if (last_warning != now) {
last_warning = now;
cmn_err(CE_WARN, "tmp_memalloc: tmpfs over memory limit");
}
return (NULL);
}
void
__zfs_dbgmsg(char *buf)
{
zfs_dbgmsg_t *zdm;
int size;
size = sizeof (zfs_dbgmsg_t) + strlen(buf);
zdm = kmem_zalloc(size, KM_SLEEP);
zdm->zdm_size = size;
zdm->zdm_timestamp = gethrestime_sec();
strcpy(zdm->zdm_msg, buf);
mutex_enter(&zfs_dbgmsgs_lock);
list_insert_tail(&zfs_dbgmsgs, zdm);
zfs_dbgmsg_size += size;
zfs_dbgmsg_purge(MAX(zfs_dbgmsg_maxsize, 0));
mutex_exit(&zfs_dbgmsgs_lock);
}
static void
dsl_crypto_key_new_sync(void *arg1, dmu_tx_t *tx)
{
struct knarg *kn = arg1;
dsl_dataset_t *ds = kn->kn_ds;
/*
* Generate a new key and add it to the keychain to be valid from
* this txg onwards.
*/
dsl_keychain_set_key(ds->ds_dir, tx, kn->kn_wkeybuf, kn->kn_wkeylen,
gethrestime_sec());
zcrypt_keychain_insert(&kn->kn_skn->skn_keychain,
tx->tx_txg, kn->kn_txgkey);
spa_history_log_internal(
dsl_dataset_get_spa(ds), "key create", tx,
"rekey succeeded dataset = %llu", ds->ds_object);
}
/*
* Update the uberblock and return TRUE if anything changed in this
* transaction group.
*/
boolean_t
uberblock_update(uberblock_t *ub, vdev_t *rvd, uint64_t txg, uint64_t mmp_delay)
{
ASSERT(ub->ub_txg < txg);
/*
* We explicitly do not set ub_version here, so that older versions
* continue to be written with the previous uberblock version.
*/
ub->ub_magic = UBERBLOCK_MAGIC;
ub->ub_txg = txg;
ub->ub_guid_sum = rvd->vdev_guid_sum;
ub->ub_timestamp = gethrestime_sec();
ub->ub_software_version = SPA_VERSION;
ub->ub_mmp_magic = MMP_MAGIC;
ub->ub_mmp_delay = spa_multihost(rvd->vdev_spa) ? mmp_delay : 0;
ub->ub_mmp_seq = 0;
return (ub->ub_rootbp.blk_birth == txg);
}
/*
* Create a pipe end by...
* allocating a vnode-fifonode pair and initializing the fifonode.
*/
void
makepipe(vnode_t **vpp1, vnode_t **vpp2)
{
fifonode_t *fnp1;
fifonode_t *fnp2;
vnode_t *nvp1;
vnode_t *nvp2;
fifodata_t *fdp;
time_t now;
fdp = kmem_cache_alloc(pipe_cache, KM_SLEEP);
fdp->fifo_lock.flk_ref = 2;
fnp1 = &fdp->fifo_fnode[0];
fnp2 = &fdp->fifo_fnode[1];
fnp1->fn_wcnt = fnp2->fn_wcnt = 1;
fnp1->fn_rcnt = fnp2->fn_rcnt = 1;
#if FIFODEBUG
if (! Fifo_fastmode) {
fnp1->fn_flag = fnp2->fn_flag = ISPIPE;
} else {
fnp1->fn_flag = fnp2->fn_flag = ISPIPE | FIFOFAST;
}
#else /* FIFODEBUG */
fnp1->fn_flag = fnp2->fn_flag = ISPIPE | FIFOFAST;
#endif /* FIFODEBUG */
now = gethrestime_sec();
fnp1->fn_atime = fnp2->fn_atime = now;
fnp1->fn_mtime = fnp2->fn_mtime = now;
fnp1->fn_ctime = fnp2->fn_ctime = now;
*vpp1 = nvp1 = FTOV(fnp1);
*vpp2 = nvp2 = FTOV(fnp2);
fifo_reinit_vp(nvp1); /* Reinitialize vnodes for reuse... */
fifo_reinit_vp(nvp2);
nvp1->v_vfsp = fifovfsp; /* Need to re-establish VFS & device */
nvp2->v_vfsp = fifovfsp; /* before we can reuse this vnode. */
nvp1->v_rdev = fifodev;
nvp2->v_rdev = fifodev;
}
/*
* The caller is done with the callback info. It may be that the
* caller's RPC failed and the NFSv4 client has actually provided new
* callback information. If so, let the caller know so they can
* advantage of this and maybe retry the RPC that originally failed.
*/
static int
rfs4_cbinfo_rele(rfs4_cbinfo_t *cbp, rfs4_cbstate_t newstate)
{
int cb_new = FALSE;
mutex_enter(cbp->cb_lock);
/* The caller gets a chance to mark the callback info as bad */
if (newstate != CB_NOCHANGE)
cbp->cb_state = newstate;
if (newstate == CB_FAILED) {
cbp->cb_timefailed = gethrestime_sec(); /* observability */
cbp->cb_notified_of_cb_path_down = FALSE;
}
cbp->cb_refcnt--; /* no longer using the information */
/*
* A thread may be waiting on this one to finish and if so,
* let it know that it is okay to do the CB_NULL to the
* client's callback server.
*/
if (cbp->cb_refcnt == 0 && cbp->cb_nullcaller)
cv_broadcast(cbp->cb_cv_nullcaller);
/*
* If this is the last thread to use the callback info and
* there is new callback information to try and no thread is
* there ready to do the CB_NULL, then return true to teh
* caller so they can do the CB_NULL
*/
if (cbp->cb_refcnt == 0 &&
cbp->cb_nullcaller == FALSE &&
cbp->cb_newer.cb_new == TRUE &&
cbp->cb_newer.cb_confirmed == TRUE)
cb_new = TRUE;
mutex_exit(cbp->cb_lock);
return (cb_new);
}
/*
* Given the desired delegation type and the "history" of the file
* determine the actual delegation type to return.
*/
static open_delegation_type4
rfs4_delegation_policy(open_delegation_type4 dtype,
rfs4_dinfo_t *dinfo, clientid4 cid)
{
time_t elapsed;
if (rfs4_deleg_policy != SRV_NORMAL_DELEGATE)
return (OPEN_DELEGATE_NONE);
/*
* Has this file/delegation ever been recalled? If not then
* no further checks for a delegation race need to be done.
* However if a recall has occurred, then check to see if a
* client has caused its own delegation recall to occur. If
* not, then has a delegation for this file been returned
* recently? If so, then do not assign a new delegation to
* avoid a "delegation race" between the original client and
* the new/conflicting client.
*/
if (dinfo->rd_ever_recalled == TRUE) {
if (dinfo->rd_conflicted_client != cid) {
elapsed = gethrestime_sec() - dinfo->rd_time_returned;
if (elapsed < rfs4_lease_time)
return (OPEN_DELEGATE_NONE);
}
}
/* Limit the number of read grants */
if (dtype == OPEN_DELEGATE_READ &&
dinfo->rd_rdgrants > MAX_READ_DELEGATIONS)
return (OPEN_DELEGATE_NONE);
/*
* Should consider limiting total number of read/write
* delegations the server will permit.
*/
return (dtype);
}
/*
* Note: ci_numusers should be the number of users connected to
* the share rather than the number of references on the tree but
* we don't have a mechanism to track users/share in smbsrv yet.
*/
static void
smb_tree_netinfo_init(smb_tree_t *tree, smb_netconnectinfo_t *info)
{
smb_user_t *user;
ASSERT(tree);
info->ci_id = tree->t_tid;
info->ci_type = tree->t_res_type;
info->ci_numopens = tree->t_open_files;
info->ci_numusers = tree->t_refcnt;
info->ci_time = gethrestime_sec() - tree->t_connect_time;
info->ci_sharelen = strlen(tree->t_sharename) + 1;
info->ci_share = smb_strdup(tree->t_sharename);
user = tree->t_user;
ASSERT(user);
info->ci_namelen = user->u_domain_len + user->u_name_len + 2;
info->ci_username = kmem_alloc(info->ci_namelen, KM_SLEEP);
(void) snprintf(info->ci_username, info->ci_namelen, "%s\\%s",
user->u_domain, user->u_name);
}
static void
rfs4_recall_file(rfs4_file_t *fp,
void (*recall)(rfs4_deleg_state_t *, bool_t trunc),
bool_t trunc, rfs4_client_t *cp)
{
struct master_recall_args *args;
rfs4_dbe_lock(fp->rf_dbe);
if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
rfs4_dbe_unlock(fp->rf_dbe);
return;
}
rfs4_dbe_hold(fp->rf_dbe); /* hold for new thread */
/*
* Mark the time we started the recall processing.
* If it has been previously recalled, do not reset the
* timer since this is used for the revocation decision.
*/
if (fp->rf_dinfo.rd_time_recalled == 0)
fp->rf_dinfo.rd_time_recalled = gethrestime_sec();
fp->rf_dinfo.rd_ever_recalled = TRUE; /* used for policy decision */
/* Client causing recall not always available */
if (cp)
fp->rf_dinfo.rd_conflicted_client = cp->rc_clientid;
rfs4_dbe_unlock(fp->rf_dbe);
args = kmem_alloc(sizeof (struct master_recall_args), KM_SLEEP);
args->fp = fp;
args->recall = recall;
args->trunc = trunc;
(void) thread_create(NULL, 0, do_recall_file, args, 0, &p0, TS_RUN,
minclsyspri);
}
/*ARGSUSED*/
static void
spa_history_log_sync(void *arg1, void *arg2, dmu_tx_t *tx)
{
spa_t *spa = arg1;
history_arg_t *hap = arg2;
const char *history_str = hap->ha_history_str;
objset_t *mos = spa->spa_meta_objset;
dmu_buf_t *dbp;
spa_history_phys_t *shpp;
size_t reclen;
uint64_t le_len;
nvlist_t *nvrecord;
char *record_packed = NULL;
int ret;
/*
* If we have an older pool that doesn't have a command
* history object, create it now.
*/
mutex_enter(&spa->spa_history_lock);
if (!spa->spa_history)
spa_history_create_obj(spa, tx);
mutex_exit(&spa->spa_history_lock);
/*
* Get the offset of where we need to write via the bonus buffer.
* Update the offset when the write completes.
*/
VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
shpp = dbp->db_data;
dmu_buf_will_dirty(dbp, tx);
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbp, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
}
#endif
VERIFY(nvlist_alloc(&nvrecord, NV_UNIQUE_NAME, KM_SLEEP) == 0);
VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TIME,
gethrestime_sec()) == 0);
VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_WHO, hap->ha_uid) == 0);
if (hap->ha_zone != NULL)
VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_ZONE,
hap->ha_zone) == 0);
#ifdef _KERNEL
VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_HOST,
utsname.nodename) == 0);
#endif
if (hap->ha_log_type == LOG_CMD_POOL_CREATE ||
hap->ha_log_type == LOG_CMD_NORMAL) {
VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_CMD,
history_str) == 0);
zfs_dbgmsg("command: %s", history_str);
} else {
VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_INT_EVENT,
hap->ha_event) == 0);
VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TXG,
tx->tx_txg) == 0);
VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_INT_STR,
history_str) == 0);
zfs_dbgmsg("internal %s pool:%s txg:%llu %s",
zfs_history_event_names[hap->ha_event], spa_name(spa),
(longlong_t)tx->tx_txg, history_str);
}
VERIFY(nvlist_size(nvrecord, &reclen, NV_ENCODE_XDR) == 0);
record_packed = kmem_alloc(reclen, KM_SLEEP);
VERIFY(nvlist_pack(nvrecord, &record_packed, &reclen,
NV_ENCODE_XDR, KM_SLEEP) == 0);
mutex_enter(&spa->spa_history_lock);
if (hap->ha_log_type == LOG_CMD_POOL_CREATE)
VERIFY(shpp->sh_eof == shpp->sh_pool_create_len);
/* write out the packed length as little endian */
le_len = LE_64((uint64_t)reclen);
ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
if (!ret)
ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
if (!ret && hap->ha_log_type == LOG_CMD_POOL_CREATE) {
shpp->sh_pool_create_len += sizeof (le_len) + reclen;
shpp->sh_bof = shpp->sh_pool_create_len;
}
mutex_exit(&spa->spa_history_lock);
nvlist_free(nvrecord);
kmem_free(record_packed, reclen);
dmu_buf_rele(dbp, FTAG);
strfree(hap->ha_history_str);
if (hap->ha_zone != NULL)
strfree(hap->ha_zone);
kmem_free(hap, sizeof (history_arg_t));
}
/*ARGSUSED*/
static void
spa_history_log_sync(void *arg, dmu_tx_t *tx)
{
nvlist_t *nvl = arg;
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
objset_t *mos = spa->spa_meta_objset;
dmu_buf_t *dbp;
spa_history_phys_t *shpp;
size_t reclen;
uint64_t le_len;
char *record_packed = NULL;
int ret;
/*
* If we have an older pool that doesn't have a command
* history object, create it now.
*/
mutex_enter(&spa->spa_history_lock);
if (!spa->spa_history)
spa_history_create_obj(spa, tx);
mutex_exit(&spa->spa_history_lock);
/*
* Get the offset of where we need to write via the bonus buffer.
* Update the offset when the write completes.
*/
VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
shpp = dbp->db_data;
dmu_buf_will_dirty(dbp, tx);
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbp, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
}
#endif
fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
#ifdef _KERNEL
fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname.nodename);
#endif
if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
zfs_dbgmsg("command: %s",
fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
} else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
} else {
zfs_dbgmsg("txg %lld %s %s",
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
}
} else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
zfs_dbgmsg("ioctl %s",
fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
}
record_packed = fnvlist_pack(nvl, &reclen);
mutex_enter(&spa->spa_history_lock);
/* write out the packed length as little endian */
le_len = LE_64((uint64_t)reclen);
ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
if (!ret)
ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
/* The first command is the create, which we keep forever */
if (ret == 0 && shpp->sh_pool_create_len == 0 &&
nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
}
mutex_exit(&spa->spa_history_lock);
fnvlist_pack_free(record_packed, reclen);
dmu_buf_rele(dbp, FTAG);
fnvlist_free(nvl);
}
/* ARGSUSED */
static void
dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx)
{
static const char *old_names[] = {
"scrub_bookmark",
"scrub_ddt_bookmark",
"scrub_ddt_class_max",
"scrub_queue",
"scrub_min_txg",
"scrub_max_txg",
"scrub_func",
"scrub_errors",
NULL
};
dsl_pool_t *dp = scn->scn_dp;
spa_t *spa = dp->dp_spa;
int i;
/* Remove any remnants of an old-style scrub. */
for (i = 0; old_names[i]; i++) {
(void) zap_remove(dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx);
}
if (scn->scn_phys.scn_queue_obj != 0) {
VERIFY(0 == dmu_object_free(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, tx));
scn->scn_phys.scn_queue_obj = 0;
}
/*
* If we were "restarted" from a stopped state, don't bother
* with anything else.
*/
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
if (complete)
scn->scn_phys.scn_state = DSS_FINISHED;
else
scn->scn_phys.scn_state = DSS_CANCELED;
spa_history_log_internal(spa, "scan done", tx,
"complete=%u", complete);
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
mutex_enter(&spa->spa_scrub_lock);
while (spa->spa_scrub_inflight > 0) {
cv_wait(&spa->spa_scrub_io_cv,
&spa->spa_scrub_lock);
}
mutex_exit(&spa->spa_scrub_lock);
spa->spa_scrub_started = B_FALSE;
spa->spa_scrub_active = B_FALSE;
/*
* If the scrub/resilver completed, update all DTLs to
* reflect this. Whether it succeeded or not, vacate
* all temporary scrub DTLs.
*/
vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg,
complete ? scn->scn_phys.scn_max_txg : 0, B_TRUE);
if (complete) {
spa_event_notify(spa, NULL, scn->scn_phys.scn_min_txg ?
ESC_ZFS_RESILVER_FINISH : ESC_ZFS_SCRUB_FINISH);
}
spa_errlog_rotate(spa);
/*
* We may have finished replacing a device.
* Let the async thread assess this and handle the detach.
*/
spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
}
scn->scn_phys.scn_end_time = gethrestime_sec();
}
static void
dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
pool_scan_func_t *funcp = arg;
dmu_object_type_t ot = 0;
dsl_pool_t *dp = scn->scn_dp;
spa_t *spa = dp->dp_spa;
ASSERT(scn->scn_phys.scn_state != DSS_SCANNING);
ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS);
bzero(&scn->scn_phys, sizeof (scn->scn_phys));
scn->scn_phys.scn_func = *funcp;
scn->scn_phys.scn_state = DSS_SCANNING;
scn->scn_phys.scn_min_txg = 0;
scn->scn_phys.scn_max_txg = tx->tx_txg;
scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */
scn->scn_phys.scn_start_time = gethrestime_sec();
scn->scn_phys.scn_errors = 0;
scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc;
scn->scn_restart_txg = 0;
spa_scan_stat_init(spa);
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max;
/* rewrite all disk labels */
vdev_config_dirty(spa->spa_root_vdev);
if (vdev_resilver_needed(spa->spa_root_vdev,
&scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) {
spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_START);
} else {
spa_event_notify(spa, NULL, ESC_ZFS_SCRUB_START);
}
spa->spa_scrub_started = B_TRUE;
/*
* If this is an incremental scrub, limit the DDT scrub phase
* to just the auto-ditto class (for correctness); the rest
* of the scrub should go faster using top-down pruning.
*/
if (scn->scn_phys.scn_min_txg > TXG_INITIAL)
scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO;
}
/* back to the generic stuff */
if (dp->dp_blkstats == NULL) {
dp->dp_blkstats =
kmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP);
}
bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
if (spa_version(spa) < SPA_VERSION_DSL_SCRUB)
ot = DMU_OT_ZAP_OTHER;
scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset,
ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx);
dsl_scan_sync_state(scn, tx);
spa_history_log_internal(spa, "scan setup", tx,
"func=%u mintxg=%llu maxtxg=%llu",
*funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg);
}
/*
* tcp_time_wait_processing() handles processing of incoming packets when
* the tcp_t is in the TIME_WAIT state.
*
* A TIME_WAIT tcp_t that has an associated open TCP end point (not in
* detached state) is never put on the time wait list.
*/
void
tcp_time_wait_processing(tcp_t *tcp, mblk_t *mp, uint32_t seg_seq,
uint32_t seg_ack, int seg_len, tcpha_t *tcpha, ip_recv_attr_t *ira)
{
int32_t bytes_acked;
int32_t gap;
int32_t rgap;
tcp_opt_t tcpopt;
uint_t flags;
uint32_t new_swnd = 0;
conn_t *nconnp;
conn_t *connp = tcp->tcp_connp;
tcp_stack_t *tcps = tcp->tcp_tcps;
BUMP_LOCAL(tcp->tcp_ibsegs);
DTRACE_PROBE2(tcp__trace__recv, mblk_t *, mp, tcp_t *, tcp);
flags = (unsigned int)tcpha->tha_flags & 0xFF;
new_swnd = ntohs(tcpha->tha_win) <<
((tcpha->tha_flags & TH_SYN) ? 0 : tcp->tcp_snd_ws);
if (tcp->tcp_snd_ts_ok) {
if (!tcp_paws_check(tcp, tcpha, &tcpopt)) {
tcp_xmit_ctl(NULL, tcp, tcp->tcp_snxt,
tcp->tcp_rnxt, TH_ACK);
goto done;
}
}
gap = seg_seq - tcp->tcp_rnxt;
rgap = tcp->tcp_rwnd - (gap + seg_len);
if (gap < 0) {
TCPS_BUMP_MIB(tcps, tcpInDataDupSegs);
TCPS_UPDATE_MIB(tcps, tcpInDataDupBytes,
(seg_len > -gap ? -gap : seg_len));
seg_len += gap;
if (seg_len < 0 || (seg_len == 0 && !(flags & TH_FIN))) {
if (flags & TH_RST) {
goto done;
}
if ((flags & TH_FIN) && seg_len == -1) {
/*
* When TCP receives a duplicate FIN in
* TIME_WAIT state, restart the 2 MSL timer.
* See page 73 in RFC 793. Make sure this TCP
* is already on the TIME_WAIT list. If not,
* just restart the timer.
*/
if (TCP_IS_DETACHED(tcp)) {
if (tcp_time_wait_remove(tcp, NULL) ==
B_TRUE) {
tcp_time_wait_append(tcp);
TCP_DBGSTAT(tcps,
tcp_rput_time_wait);
}
} else {
ASSERT(tcp != NULL);
TCP_TIMER_RESTART(tcp,
tcps->tcps_time_wait_interval);
}
tcp_xmit_ctl(NULL, tcp, tcp->tcp_snxt,
tcp->tcp_rnxt, TH_ACK);
goto done;
}
flags |= TH_ACK_NEEDED;
seg_len = 0;
goto process_ack;
}
/* Fix seg_seq, and chew the gap off the front. */
seg_seq = tcp->tcp_rnxt;
}
if ((flags & TH_SYN) && gap > 0 && rgap < 0) {
/*
* Make sure that when we accept the connection, pick
* an ISS greater than (tcp_snxt + ISS_INCR/2) for the
* old connection.
*
* The next ISS generated is equal to tcp_iss_incr_extra
* + ISS_INCR/2 + other components depending on the
* value of tcp_strong_iss. We pre-calculate the new
* ISS here and compare with tcp_snxt to determine if
* we need to make adjustment to tcp_iss_incr_extra.
*
* The above calculation is ugly and is a
* waste of CPU cycles...
*/
uint32_t new_iss = tcps->tcps_iss_incr_extra;
int32_t adj;
ip_stack_t *ipst = tcps->tcps_netstack->netstack_ip;
switch (tcps->tcps_strong_iss) {
case 2: {
/* Add time and MD5 components. */
uint32_t answer[4];
struct {
uint32_t ports;
in6_addr_t src;
in6_addr_t dst;
} arg;
MD5_CTX context;
mutex_enter(&tcps->tcps_iss_key_lock);
context = tcps->tcps_iss_key;
mutex_exit(&tcps->tcps_iss_key_lock);
arg.ports = connp->conn_ports;
/* We use MAPPED addresses in tcp_iss_init */
arg.src = connp->conn_laddr_v6;
arg.dst = connp->conn_faddr_v6;
MD5Update(&context, (uchar_t *)&arg,
sizeof (arg));
MD5Final((uchar_t *)answer, &context);
answer[0] ^= answer[1] ^ answer[2] ^ answer[3];
new_iss += (gethrtime() >> ISS_NSEC_SHT) + answer[0];
break;
}
case 1:
/* Add time component and min random (i.e. 1). */
new_iss += (gethrtime() >> ISS_NSEC_SHT) + 1;
break;
default:
/* Add only time component. */
new_iss += (uint32_t)gethrestime_sec() * ISS_INCR;
break;
}
if ((adj = (int32_t)(tcp->tcp_snxt - new_iss)) > 0) {
/*
* New ISS not guaranteed to be ISS_INCR/2
* ahead of the current tcp_snxt, so add the
* difference to tcp_iss_incr_extra.
*/
tcps->tcps_iss_incr_extra += adj;
}
/*
* If tcp_clean_death() can not perform the task now,
* drop the SYN packet and let the other side re-xmit.
* Otherwise pass the SYN packet back in, since the
* old tcp state has been cleaned up or freed.
*/
if (tcp_clean_death(tcp, 0) == -1)
goto done;
nconnp = ipcl_classify(mp, ira, ipst);
if (nconnp != NULL) {
TCP_STAT(tcps, tcp_time_wait_syn_success);
/* Drops ref on nconnp */
tcp_reinput(nconnp, mp, ira, ipst);
return;
}
goto done;
}
/*
* msgctl system call.
*
* gets q lock (via ipc_lookup), releases before return.
* may call users of msg_lock
*/
static int
msgctl(int msgid, int cmd, void *arg)
{
STRUCT_DECL(msqid_ds, ds); /* SVR4 queue work area */
kmsqid_t *qp; /* ptr to associated q */
int error;
struct cred *cr;
model_t mdl = get_udatamodel();
struct msqid_ds64 ds64;
kmutex_t *lock;
proc_t *pp = curproc;
STRUCT_INIT(ds, mdl);
cr = CRED();
/*
* Perform pre- or non-lookup actions (e.g. copyins, RMID).
*/
switch (cmd) {
case IPC_SET:
if (copyin(arg, STRUCT_BUF(ds), STRUCT_SIZE(ds)))
return (set_errno(EFAULT));
break;
case IPC_SET64:
if (copyin(arg, &ds64, sizeof (struct msqid_ds64)))
return (set_errno(EFAULT));
break;
case IPC_RMID:
if (error = ipc_rmid(msq_svc, msgid, cr))
return (set_errno(error));
return (0);
}
/*
* get msqid_ds for this msgid
*/
if ((lock = ipc_lookup(msq_svc, msgid, (kipc_perm_t **)&qp)) == NULL)
return (set_errno(EINVAL));
switch (cmd) {
case IPC_SET:
if (STRUCT_FGET(ds, msg_qbytes) > qp->msg_qbytes &&
secpolicy_ipc_config(cr) != 0) {
mutex_exit(lock);
return (set_errno(EPERM));
}
if (error = ipcperm_set(msq_svc, cr, &qp->msg_perm,
&STRUCT_BUF(ds)->msg_perm, mdl)) {
mutex_exit(lock);
return (set_errno(error));
}
qp->msg_qbytes = STRUCT_FGET(ds, msg_qbytes);
qp->msg_ctime = gethrestime_sec();
break;
case IPC_STAT:
if (error = ipcperm_access(&qp->msg_perm, MSG_R, cr)) {
mutex_exit(lock);
return (set_errno(error));
}
if (qp->msg_rcv_cnt)
qp->msg_perm.ipc_mode |= MSG_RWAIT;
if (qp->msg_snd_cnt)
qp->msg_perm.ipc_mode |= MSG_WWAIT;
ipcperm_stat(&STRUCT_BUF(ds)->msg_perm, &qp->msg_perm, mdl);
qp->msg_perm.ipc_mode &= ~(MSG_RWAIT|MSG_WWAIT);
STRUCT_FSETP(ds, msg_first, NULL); /* kernel addr */
STRUCT_FSETP(ds, msg_last, NULL);
STRUCT_FSET(ds, msg_cbytes, qp->msg_cbytes);
STRUCT_FSET(ds, msg_qnum, qp->msg_qnum);
STRUCT_FSET(ds, msg_qbytes, qp->msg_qbytes);
STRUCT_FSET(ds, msg_lspid, qp->msg_lspid);
STRUCT_FSET(ds, msg_lrpid, qp->msg_lrpid);
STRUCT_FSET(ds, msg_stime, qp->msg_stime);
STRUCT_FSET(ds, msg_rtime, qp->msg_rtime);
STRUCT_FSET(ds, msg_ctime, qp->msg_ctime);
break;
case IPC_SET64:
mutex_enter(&pp->p_lock);
if ((ds64.msgx_qbytes > qp->msg_qbytes) &&
secpolicy_ipc_config(cr) != 0 &&
rctl_test(rc_process_msgmnb, pp->p_rctls, pp,
ds64.msgx_qbytes, RCA_SAFE) & RCT_DENY) {
mutex_exit(&pp->p_lock);
mutex_exit(lock);
return (set_errno(EPERM));
}
mutex_exit(&pp->p_lock);
if (error = ipcperm_set64(msq_svc, cr, &qp->msg_perm,
&ds64.msgx_perm)) {
mutex_exit(lock);
return (set_errno(error));
}
qp->msg_qbytes = ds64.msgx_qbytes;
qp->msg_ctime = gethrestime_sec();
break;
case IPC_STAT64:
if (qp->msg_rcv_cnt)
qp->msg_perm.ipc_mode |= MSG_RWAIT;
if (qp->msg_snd_cnt)
qp->msg_perm.ipc_mode |= MSG_WWAIT;
ipcperm_stat64(&ds64.msgx_perm, &qp->msg_perm);
qp->msg_perm.ipc_mode &= ~(MSG_RWAIT|MSG_WWAIT);
ds64.msgx_cbytes = qp->msg_cbytes;
ds64.msgx_qnum = qp->msg_qnum;
ds64.msgx_qbytes = qp->msg_qbytes;
ds64.msgx_lspid = qp->msg_lspid;
ds64.msgx_lrpid = qp->msg_lrpid;
ds64.msgx_stime = qp->msg_stime;
ds64.msgx_rtime = qp->msg_rtime;
ds64.msgx_ctime = qp->msg_ctime;
break;
default:
mutex_exit(lock);
return (set_errno(EINVAL));
}
mutex_exit(lock);
/*
* Do copyout last (after releasing mutex).
*/
switch (cmd) {
case IPC_STAT:
if (copyout(STRUCT_BUF(ds), arg, STRUCT_SIZE(ds)))
return (set_errno(EFAULT));
break;
case IPC_STAT64:
if (copyout(&ds64, arg, sizeof (struct msqid_ds64)))
return (set_errno(EFAULT));
break;
}
return (0);
}
/*ARGSUSED*/
static void
spa_history_log_sync(void *arg, dmu_tx_t *tx)
{
nvlist_t *nvl = arg;
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
objset_t *mos = spa->spa_meta_objset;
dmu_buf_t *dbp;
spa_history_phys_t *shpp;
size_t reclen;
uint64_t le_len;
char *record_packed = NULL;
int ret;
/*
* If we have an older pool that doesn't have a command
* history object, create it now.
*/
mutex_enter(&spa->spa_history_lock);
if (!spa->spa_history)
spa_history_create_obj(spa, tx);
mutex_exit(&spa->spa_history_lock);
/*
* Get the offset of where we need to write via the bonus buffer.
* Update the offset when the write completes.
*/
VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
shpp = dbp->db_data;
dmu_buf_will_dirty(dbp, tx);
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbp, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
}
#endif
fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname()->nodename);
if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
zfs_dbgmsg("command: %s",
fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
} else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
} else {
zfs_dbgmsg("txg %lld %s %s",
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
}
/*
* The history sysevent is posted only for internal history
* messages to show what has happened, not how it happened. For
* example, the following command:
*
* # zfs destroy -r tank/foo
*
* will result in one sysevent posted per dataset that is
* destroyed as a result of the command - which could be more
* than one event in total. By contrast, if the sysevent was
* posted as a result of the ZPOOL_HIST_CMD key being present
* it would result in only one sysevent being posted with the
* full command line arguments, requiring the consumer to know
* how to parse and understand zfs(1M) command invocations.
*/
spa_history_log_notify(spa, nvl);
} else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
zfs_dbgmsg("ioctl %s",
fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
}
VERIFY3U(nvlist_pack(nvl, &record_packed, &reclen, NV_ENCODE_NATIVE,
KM_SLEEP), ==, 0);
mutex_enter(&spa->spa_history_lock);
/* write out the packed length as little endian */
le_len = LE_64((uint64_t)reclen);
ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
if (!ret)
ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
/* The first command is the create, which we keep forever */
if (ret == 0 && shpp->sh_pool_create_len == 0 &&
nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
}
mutex_exit(&spa->spa_history_lock);
fnvlist_pack_free(record_packed, reclen);
dmu_buf_rele(dbp, FTAG);
fnvlist_free(nvl);
}
