/*
* Get exclusive access to the logging system; this includes all minor
* devices. We use an rwlock rather than a mutex because hold times
* are potentially long, so we don't want to waste cycles in adaptive mutex
* spin (rwlocks always block when contended). Note that we explicitly
* support recursive calls (e.g. printf() calls foo() calls printf()).
*
* Clients may use log_enter() / log_exit() to guarantee that a group
* of messages is treated atomically (i.e. they appear in order and are
* not interspersed with any other messages), e.g. for multiline printf().
*
* This could probably be changed to a per-zone lock if contention becomes
* an issue.
*/
void
log_enter(void)
{
if (rw_owner(&log_rwlock) != curthread)
rw_enter(&log_rwlock, RW_WRITER);
log_rwlock_depth++;
}
vnode_t *
makenfs4node_by_fh(nfs4_sharedfh_t *sfh, nfs4_sharedfh_t *psfh,
nfs4_fname_t **npp, nfs4_ga_res_t *garp,
mntinfo4_t *mi, cred_t *cr, hrtime_t t)
{
vfs_t *vfsp = mi->mi_vfsp;
int newnode = 0;
vnode_t *vp;
rnode4_t *rp;
svnode_t *svp;
nfs4_fname_t *name, *svpname;
int index;
ASSERT(npp && *npp);
name = *npp;
*npp = NULL;
index = rtable4hash(sfh);
rw_enter(&rtable4[index].r_lock, RW_READER);
vp = make_rnode4(sfh, &rtable4[index], vfsp,
nfs4_vnodeops, nfs4_putapage, &newnode, cr);
svp = VTOSV(vp);
rp = VTOR4(vp);
if (newnode) {
svp->sv_forw = svp->sv_back = svp;
svp->sv_name = name;
if (psfh != NULL)
sfh4_hold(psfh);
svp->sv_dfh = psfh;
} else {
/*
* It is possible that due to a server
* side rename fnames have changed.
* update the fname here.
*/
mutex_enter(&rp->r_svlock);
svpname = svp->sv_name;
if (svp->sv_name != name) {
svp->sv_name = name;
mutex_exit(&rp->r_svlock);
fn_rele(&svpname);
} else {
mutex_exit(&rp->r_svlock);
fn_rele(&name);
}
}
ASSERT(RW_LOCK_HELD(&rtable4[index].r_lock));
r4_do_attrcache(vp, garp, newnode, t, cr, index);
ASSERT(rw_owner(&rtable4[index].r_lock) != curthread);
return (vp);
}
/*
* Find or create a vnode for the given filehandle, filesystem, parent, and
* name. The reference to nm is consumed, so the caller must first do an
* fn_hold() if it wants to continue using nm after this call.
*/
vnode_t *
makenfs4node(nfs4_sharedfh_t *fh, nfs4_ga_res_t *garp, struct vfs *vfsp,
hrtime_t t, cred_t *cr, vnode_t *dvp, nfs4_fname_t *nm)
{
vnode_t *vp;
int newnode;
int index;
mntinfo4_t *mi = VFTOMI4(vfsp);
int had_badfh = 0;
rnode4_t *rp;
ASSERT(dvp != NULL);
fh = badrootfh_check(fh, nm, mi, &had_badfh);
index = rtable4hash(fh);
rw_enter(&rtable4[index].r_lock, RW_READER);
/*
* Note: make_rnode4() may upgrade the hash bucket lock to exclusive.
*/
vp = make_rnode4(fh, &rtable4[index], vfsp, nfs4_vnodeops,
nfs4_putapage, &newnode, cr);
rp = VTOR4(vp);
sv_activate(&vp, dvp, &nm, newnode);
if (dvp->v_flag & V_XATTRDIR) {
mutex_enter(&rp->r_statelock);
rp->r_flags |= R4ISXATTR;
mutex_exit(&rp->r_statelock);
}
/* if getting a bad file handle, do not cache the attributes. */
if (had_badfh) {
rw_exit(&rtable4[index].r_lock);
return (vp);
}
ASSERT(RW_LOCK_HELD(&rtable4[index].r_lock));
r4_do_attrcache(vp, garp, newnode, t, cr, index);
ASSERT(rw_owner(&rtable4[index].r_lock) != curthread);
return (vp);
}
/* ARGSUSED */
static void
auto_inactive(vnode_t *vp, cred_t *cred, caller_context_t *ct)
{
fnnode_t *fnp = vntofn(vp);
fnnode_t *dfnp = fnp->fn_parent;
int count;
AUTOFS_DPRINT((4, "auto_inactive: vp=%p v_count=%u fn_link=%d\n",
(void *)vp, vp->v_count, fnp->fn_linkcnt));
/*
* The rwlock should not be already held by this thread.
* The assert relies on the fact that the owner field is cleared
* when the lock is released.
*/
ASSERT(dfnp != NULL);
ASSERT(rw_owner(&dfnp->fn_rwlock) != curthread);
rw_enter(&dfnp->fn_rwlock, RW_WRITER);
mutex_enter(&vp->v_lock);
ASSERT(vp->v_count > 0);
count = --vp->v_count;
mutex_exit(&vp->v_lock);
if (count == 0) {
/*
* Free only if node has no subdirectories.
*/
if (fnp->fn_linkcnt == 1) {
auto_disconnect(dfnp, fnp);
rw_exit(&dfnp->fn_rwlock);
auto_freefnnode(fnp);
AUTOFS_DPRINT((5, "auto_inactive: (exit) vp=%p freed\n",
(void *)vp));
return;
}
}
rw_exit(&dfnp->fn_rwlock);
AUTOFS_DPRINT((5, "auto_inactive: (exit) vp=%p v_count=%u fn_link=%d\n",
(void *)vp, vp->v_count, fnp->fn_linkcnt));
}
/*
* This function is called when we are unable to obtain a write lock on the
* first try. This means that at least one other thread holds either a
* read or write lock.
*/
void
__rw_wlock_hard(volatile uintptr_t *c, uintptr_t tid, const char *file,
int line)
{
struct rwlock *rw;
struct turnstile *ts;
#ifdef ADAPTIVE_RWLOCKS
volatile struct thread *owner;
int spintries = 0;
int i;
#endif
uintptr_t v, x;
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
#ifdef KDTRACE_HOOKS
uintptr_t state;
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
int64_t all_time = 0;
#endif
if (SCHEDULER_STOPPED())
return;
rw = rwlock2rw(c);
if (rw_wlocked(rw)) {
KASSERT(rw->lock_object.lo_flags & LO_RECURSABLE,
("%s: recursing but non-recursive rw %s @ %s:%d\n",
__func__, rw->lock_object.lo_name, file, line));
rw->rw_recurse++;
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p recursing", __func__, rw);
return;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR5(KTR_LOCK, "%s: %s contested (lock=%p) at %s:%d", __func__,
rw->lock_object.lo_name, (void *)rw->rw_lock, file, line);
#ifdef KDTRACE_HOOKS
all_time -= lockstat_nsecs(&rw->lock_object);
state = rw->rw_lock;
#endif
while (!_rw_write_lock(rw, tid)) {
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&rw->lock_object,
&contested, &waittime);
#ifdef ADAPTIVE_RWLOCKS
/*
* If the lock is write locked and the owner is
* running on another CPU, spin until the owner stops
* running or the state of the lock changes.
*/
v = rw->rw_lock;
owner = (struct thread *)RW_OWNER(v);
if (!(v & RW_LOCK_READ) && TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR3(KTR_LOCK, "%s: spinning on %p held by %p",
__func__, rw, owner);
KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
"spinning", "lockname:\"%s\"",
rw->lock_object.lo_name);
while ((struct thread*)RW_OWNER(rw->rw_lock) == owner &&
TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
"running");
continue;
}
if ((v & RW_LOCK_READ) && RW_READERS(v) &&
spintries < rowner_retries) {
if (!(v & RW_LOCK_WRITE_SPINNER)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_WRITE_SPINNER)) {
continue;
}
}
spintries++;
KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
"spinning", "lockname:\"%s\"",
rw->lock_object.lo_name);
for (i = 0; i < rowner_loops; i++) {
if ((rw->rw_lock & RW_LOCK_WRITE_SPINNER) == 0)
break;
cpu_spinwait();
}
KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
"running");
#ifdef KDTRACE_HOOKS
spin_cnt += rowner_loops - i;
#endif
if (i != rowner_loops)
continue;
}
#endif
ts = turnstile_trywait(&rw->lock_object);
v = rw->rw_lock;
#ifdef ADAPTIVE_RWLOCKS
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owners) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
if (!(v & RW_LOCK_READ)) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
}
#endif
/*
* Check for the waiters flags about this rwlock.
* If the lock was released, without maintain any pending
* waiters queue, simply try to acquire it.
* If a pending waiters queue is present, claim the lock
* ownership and maintain the pending queue.
*/
x = v & (RW_LOCK_WAITERS | RW_LOCK_WRITE_SPINNER);
if ((v & ~x) == RW_UNLOCKED) {
x &= ~RW_LOCK_WRITE_SPINNER;
if (atomic_cmpset_acq_ptr(&rw->rw_lock, v, tid | x)) {
if (x)
turnstile_claim(ts);
else
turnstile_cancel(ts);
break;
}
turnstile_cancel(ts);
continue;
}
/*
* If the RW_LOCK_WRITE_WAITERS flag isn't set, then try to
* set it. If we fail to set it, then loop back and try
* again.
*/
if (!(v & RW_LOCK_WRITE_WAITERS)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_WRITE_WAITERS)) {
turnstile_cancel(ts);
continue;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p set write waiters flag",
__func__, rw);
}
/*
* We were unable to acquire the lock and the write waiters
* flag is set, so we must block on the turnstile.
*/
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p blocking on turnstile", __func__,
rw);
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs(&rw->lock_object);
#endif
turnstile_wait(ts, rw_owner(rw), TS_EXCLUSIVE_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs(&rw->lock_object);
sleep_cnt++;
#endif
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p resuming from turnstile",
__func__, rw);
#ifdef ADAPTIVE_RWLOCKS
spintries = 0;
#endif
}
#ifdef KDTRACE_HOOKS
all_time += lockstat_nsecs(&rw->lock_object);
if (sleep_time)
LOCKSTAT_RECORD4(rw__block, rw, sleep_time,
LOCKSTAT_WRITER, (state & RW_LOCK_READ) == 0,
(state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state));
/* Record only the loops spinning and not sleeping. */
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD4(rw__spin, rw, all_time - sleep_time,
LOCKSTAT_READER, (state & RW_LOCK_READ) == 0,
(state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state));
#endif
LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, contested,
waittime, file, line, LOCKSTAT_WRITER);
}
void
__rw_rlock(volatile uintptr_t *c, const char *file, int line)
{
struct rwlock *rw;
struct turnstile *ts;
#ifdef ADAPTIVE_RWLOCKS
volatile struct thread *owner;
int spintries = 0;
int i;
#endif
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
uintptr_t v;
#ifdef KDTRACE_HOOKS
uintptr_t state;
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
int64_t all_time = 0;
#endif
if (SCHEDULER_STOPPED())
return;
rw = rwlock2rw(c);
KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
("rw_rlock() by idle thread %p on rwlock %s @ %s:%d",
curthread, rw->lock_object.lo_name, file, line));
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_rlock() of destroyed rwlock @ %s:%d", file, line));
KASSERT(rw_wowner(rw) != curthread,
("rw_rlock: wlock already held for %s @ %s:%d",
rw->lock_object.lo_name, file, line));
WITNESS_CHECKORDER(&rw->lock_object, LOP_NEWORDER, file, line, NULL);
#ifdef KDTRACE_HOOKS
all_time -= lockstat_nsecs(&rw->lock_object);
state = rw->rw_lock;
#endif
for (;;) {
/*
* Handle the easy case. If no other thread has a write
* lock, then try to bump up the count of read locks. Note
* that we have to preserve the current state of the
* RW_LOCK_WRITE_WAITERS flag. If we fail to acquire a
* read lock, then rw_lock must have changed, so restart
* the loop. Note that this handles the case of a
* completely unlocked rwlock since such a lock is encoded
* as a read lock with no waiters.
*/
v = rw->rw_lock;
if (RW_CAN_READ(v)) {
/*
* The RW_LOCK_READ_WAITERS flag should only be set
* if the lock has been unlocked and write waiters
* were present.
*/
if (atomic_cmpset_acq_ptr(&rw->rw_lock, v,
v + RW_ONE_READER)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR4(KTR_LOCK,
"%s: %p succeed %p -> %p", __func__,
rw, (void *)v,
(void *)(v + RW_ONE_READER));
break;
}
continue;
}
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&rw->lock_object,
&contested, &waittime);
#ifdef ADAPTIVE_RWLOCKS
/*
* If the owner is running on another CPU, spin until
* the owner stops running or the state of the lock
* changes.
*/
if ((v & RW_LOCK_READ) == 0) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, rw, owner);
KTR_STATE1(KTR_SCHED, "thread",
sched_tdname(curthread), "spinning",
"lockname:\"%s\"", rw->lock_object.lo_name);
while ((struct thread*)RW_OWNER(rw->rw_lock) ==
owner && TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
KTR_STATE0(KTR_SCHED, "thread",
sched_tdname(curthread), "running");
continue;
}
} else if (spintries < rowner_retries) {
spintries++;
KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
"spinning", "lockname:\"%s\"",
rw->lock_object.lo_name);
for (i = 0; i < rowner_loops; i++) {
v = rw->rw_lock;
if ((v & RW_LOCK_READ) == 0 || RW_CAN_READ(v))
break;
cpu_spinwait();
}
#ifdef KDTRACE_HOOKS
spin_cnt += rowner_loops - i;
#endif
KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
"running");
if (i != rowner_loops)
continue;
}
#endif
/*
* Okay, now it's the hard case. Some other thread already
* has a write lock or there are write waiters present,
* acquire the turnstile lock so we can begin the process
* of blocking.
*/
ts = turnstile_trywait(&rw->lock_object);
/*
* The lock might have been released while we spun, so
* recheck its state and restart the loop if needed.
*/
v = rw->rw_lock;
if (RW_CAN_READ(v)) {
turnstile_cancel(ts);
continue;
}
#ifdef ADAPTIVE_RWLOCKS
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owners) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
if ((v & RW_LOCK_READ) == 0) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
}
#endif
/*
* The lock is held in write mode or it already has waiters.
*/
MPASS(!RW_CAN_READ(v));
/*
* If the RW_LOCK_READ_WAITERS flag is already set, then
* we can go ahead and block. If it is not set then try
* to set it. If we fail to set it drop the turnstile
* lock and restart the loop.
*/
if (!(v & RW_LOCK_READ_WAITERS)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_READ_WAITERS)) {
turnstile_cancel(ts);
continue;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p set read waiters flag",
__func__, rw);
}
/*
* We were unable to acquire the lock and the read waiters
* flag is set, so we must block on the turnstile.
*/
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p blocking on turnstile", __func__,
rw);
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs(&rw->lock_object);
#endif
turnstile_wait(ts, rw_owner(rw), TS_SHARED_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs(&rw->lock_object);
sleep_cnt++;
#endif
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p resuming from turnstile",
__func__, rw);
}
#ifdef KDTRACE_HOOKS
all_time += lockstat_nsecs(&rw->lock_object);
if (sleep_time)
LOCKSTAT_RECORD4(rw__block, rw, sleep_time,
LOCKSTAT_READER, (state & RW_LOCK_READ) == 0,
(state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state));
/* Record only the loops spinning and not sleeping. */
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD4(rw__spin, rw, all_time - sleep_time,
LOCKSTAT_READER, (state & RW_LOCK_READ) == 0,
(state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state));
#endif
/*
* TODO: acquire "owner of record" here. Here be turnstile dragons
* however. turnstiles don't like owners changing between calls to
* turnstile_wait() currently.
*/
LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, contested,
waittime, file, line, LOCKSTAT_READER);
LOCK_LOG_LOCK("RLOCK", &rw->lock_object, 0, 0, file, line);
WITNESS_LOCK(&rw->lock_object, 0, file, line);
curthread->td_locks++;
curthread->td_rw_rlocks++;
}