195 Need replacement for nfs/lockd+klm

[unleashed.git] / usr / src / uts / common / klm / nlm_client.c

/*
 * Copyright (c) 2008 Isilon Inc http://www.isilon.com/
 * Authors: Doug Rabson <dfr@rabson.org>
 * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

/*
 * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
 * Copyright (c) 2012 by Delphix. All rights reserved.
 */

/*
 * Client-side support for (NFS) VOP_FRLOCK, VOP_SHRLOCK.
 * (called via klmops.c: lm_frlock, lm4_frlock)
 *
 * Source code derived from FreeBSD nlm_advlock.c
 */

#include <sys/param.h>
#include <sys/fcntl.h>
#include <sys/lock.h>
#include <sys/flock.h>
#include <sys/mount.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/share.h>
#include <sys/syslog.h>
#include <sys/systm.h>
#include <sys/unistd.h>
#include <sys/vnode.h>
#include <sys/queue.h>
#include <sys/sdt.h>
#include <netinet/in.h>

#include <fs/fs_subr.h>
#include <rpcsvc/nlm_prot.h>

#include <nfs/nfs.h>
#include <nfs/nfs_clnt.h>
#include <nfs/export.h>
#include <nfs/rnode.h>
#include <nfs/lm.h>

#include "nlm_impl.h"

/* Extra flags for nlm_call_lock() - xflags */
#define NLM_X_RECLAIM   1
#define NLM_X_BLOCKING  2

/*
 * Max. number of retries nlm_call_cancel() does
 * when NLM server is in grace period or doesn't
 * respond correctly.
 */
#define NLM_CANCEL_NRETRS 5

/*
 * Determines wether given lock "flp" is safe.
 * The lock is considered to be safe when it
 * acquires the whole file (i.e. its start
 * and len are zeroes).
 */
#define NLM_FLOCK_IS_SAFE(flp) \
        ((flp)->l_start == 0 && (flp)->l_len == 0)

static volatile uint32_t nlm_xid = 1;

static int nlm_init_fh_by_vp(vnode_t *, struct netobj *, rpcvers_t *);
static int nlm_map_status(nlm4_stats);
static int nlm_map_clnt_stat(enum clnt_stat);
static void nlm_send_siglost(pid_t);

static int nlm_frlock_getlk(struct nlm_host *, vnode_t *,
    struct flock64 *, int, u_offset_t, struct netobj *, int);

static int nlm_frlock_setlk(struct nlm_host *, vnode_t *,
    struct flock64 *, int, u_offset_t, struct netobj *,
    struct flk_callback *, int, bool_t);

static int nlm_reclaim_lock(struct nlm_host *, vnode_t *,
    struct flock64 *, int32_t);

static void nlm_init_lock(struct nlm4_lock *,
    const struct flock64 *, struct netobj *,
    struct nlm_owner_handle *);

static int nlm_call_lock(vnode_t *, struct flock64 *,
    struct nlm_host *, struct netobj *,
    struct flk_callback *, int, int);
static int nlm_call_unlock(struct flock64 *, struct nlm_host *,
    struct netobj *, int);
static int nlm_call_test(struct flock64 *, struct nlm_host *,
    struct netobj *, int);
static int nlm_call_cancel(struct nlm4_lockargs *,
    struct nlm_host *, int);

static int nlm_local_getlk(vnode_t *, struct flock64 *, int);
static int nlm_local_setlk(vnode_t *, struct flock64 *, int);
static void nlm_local_cancelk(vnode_t *, struct flock64 *);

static void nlm_init_share(struct nlm4_share *,
    const struct shrlock *, struct netobj *);

static int nlm_call_share(struct shrlock *, struct nlm_host *,
    struct netobj *, int, int);
static int nlm_call_unshare(struct shrlock *, struct nlm_host *,
    struct netobj *, int);
static int nlm_reclaim_share(struct nlm_host *, vnode_t *,
    struct shrlock *, uint32_t);
static int nlm_local_shrlock(vnode_t *, struct shrlock *, int, int);
static void nlm_local_shrcancel(vnode_t *, struct shrlock *);

/*
 * Reclaim locks/shares acquired by the client side
 * on the given server represented by hostp.
 * The function is called from a dedicated thread
 * when server reports us that it's entered grace
 * period.
 */
void
nlm_reclaim_client(struct nlm_globals *g, struct nlm_host *hostp)
{
        int32_t state;
        int error, sysid;
        struct locklist *llp_head, *llp;
        struct nlm_shres *nsp_head, *nsp;
        bool_t restart;

        sysid = hostp->nh_sysid | LM_SYSID_CLIENT;
        do {
                error = 0;
                restart = FALSE;
                state = nlm_host_get_state(hostp);

                DTRACE_PROBE3(reclaim__iter, struct nlm_globals *, g,
                    struct nlm_host *, hostp, int, state);

                /*
                 * We cancel all sleeping locks that were
                 * done by the host, because we don't allow
                 * reclamation of sleeping locks. The reason
                 * we do this is that allowing of sleeping locks
                 * reclamation can potentially break locks recovery
                 * order.
                 *
                 * Imagine that we have two client machines A and B
                 * and an NLM server machine. A adds a non sleeping
                 * lock to the file F and aquires this file. Machine
                 * B in its turn adds sleeping lock to the file
                 * F and blocks because F is already aquired by
                 * the machine A. Then server crashes and after the
                 * reboot it notifies its clients about the crash.
                 * If we would allow sleeping locks reclamation,
                 * there would be possible that machine B recovers
                 * its lock faster than machine A (by some reason).
                 * So that B aquires the file F after server crash and
                 * machine A (that by some reason recovers slower) fails
                 * to recover its non sleeping lock. Thus the original
                 * locks order becames broken.
                 */
                nlm_host_cancel_slocks(g, hostp);

                /*
                 * Try to reclaim all active locks we have
                 */
                llp_head = llp = flk_get_active_locks(sysid, NOPID);
                while (llp != NULL) {
                        error = nlm_reclaim_lock(hostp, llp->ll_vp,
                            &llp->ll_flock, state);

                        if (error == 0) {
                                llp = llp->ll_next;
                                continue;
                        } else if (error == ERESTART) {
                                restart = TRUE;
                                break;
                        } else {
                                /*
                                 * Critical error occurred, the lock
                                 * can not be recovered, just take it away.
                                 */
                                nlm_local_cancelk(llp->ll_vp, &llp->ll_flock);
                        }

                        llp = llp->ll_next;
                }

                flk_free_locklist(llp_head);
                if (restart) {
                        /*
                         * Lock reclamation fucntion reported us that
                         * the server state was changed (again), so
                         * try to repeat the whole reclamation process.
                         */
                        continue;
                }

                nsp_head = nsp = nlm_get_active_shres(hostp);
                while (nsp != NULL) {
                        error = nlm_reclaim_share(hostp, nsp->ns_vp,
                            nsp->ns_shr, state);

                        if (error == 0) {
                                nsp = nsp->ns_next;
                                continue;
                        } else if (error == ERESTART) {
                                break;
                        } else {
                                /* Failed to reclaim share */
                                nlm_shres_untrack(hostp, nsp->ns_vp,
                                    nsp->ns_shr);
                                nlm_local_shrcancel(nsp->ns_vp,
                                    nsp->ns_shr);
                        }

                        nsp = nsp->ns_next;
                }

                nlm_free_shrlist(nsp_head);
        } while (state != nlm_host_get_state(hostp));
}

/*
 * nlm_frlock --
 *      NFS advisory byte-range locks.
 *      Called in klmops.c
 *
 * Note that the local locking code (os/flock.c) is used to
 * keep track of remote locks granted by some server, so we
 * can reclaim those locks after a server restarts.  We can
 * also sometimes use this as a cache of lock information.
 *
 * Was: nlm_advlock()
 */
/* ARGSUSED */
int
nlm_frlock(struct vnode *vp, int cmd, struct flock64 *flkp,
        int flags, u_offset_t offset, struct cred *crp,
        struct netobj *fhp, struct flk_callback *flcb, int vers)
{
        mntinfo_t *mi;
        servinfo_t *sv;
        const char *netid;
        struct nlm_host *hostp;
        int error;
        struct nlm_globals *g;

        mi = VTOMI(vp);
        sv = mi->mi_curr_serv;

        netid = nlm_knc_to_netid(sv->sv_knconf);
        if (netid == NULL) {
                NLM_ERR("nlm_frlock: unknown NFS netid");
                return (ENOSYS);
        }

        g = zone_getspecific(nlm_zone_key, curzone);
        hostp = nlm_host_findcreate(g, sv->sv_hostname, netid, &sv->sv_addr);
        if (hostp == NULL)
                return (ENOSYS);

        /*
         * Purge cached attributes in order to make sure that
         * future calls of convoff()/VOP_GETATTR() will get the
         * latest data.
         */
        if (flkp->l_whence == SEEK_END)
                PURGE_ATTRCACHE(vp);

        /* Now flk0 is the zero-based lock request. */
        switch (cmd) {
        case F_GETLK:
                error = nlm_frlock_getlk(hostp, vp, flkp, flags,
                    offset, fhp, vers);
                break;

        case F_SETLK:
        case F_SETLKW:
                error = nlm_frlock_setlk(hostp, vp, flkp, flags,
                    offset, fhp, flcb, vers, (cmd == F_SETLKW));
                if (error == 0)
                        nlm_host_monitor(g, hostp, 0);
                break;

        default:
                error = EINVAL;
                break;
        }

        nlm_host_release(g, hostp);
        return (error);
}

static int
nlm_frlock_getlk(struct nlm_host *hostp, vnode_t *vp,
    struct flock64 *flkp, int flags, u_offset_t offset,
    struct netobj *fhp, int vers)
{
        struct flock64 flk0;
        int error;

        /*
         * Check local (cached) locks first.
         * If we find one, no need for RPC.
         */
        flk0 = *flkp;
        flk0.l_pid = curproc->p_pid;
        error = nlm_local_getlk(vp, &flk0, flags);
        if (error != 0)
                return (error);
        if (flk0.l_type != F_UNLCK) {
                *flkp = flk0;
                return (0);
        }

        /* Not found locally.  Try remote. */
        flk0 = *flkp;
        flk0.l_pid = curproc->p_pid;
        error = convoff(vp, &flk0, 0, (offset_t)offset);
        if (error != 0)
                return (error);

        error = nlm_call_test(&flk0, hostp, fhp, vers);
        if (error != 0)
                return (error);

        if (flk0.l_type == F_UNLCK) {
                /*
                 * Update the caller's *flkp with information
                 * on the conflicting lock (or lack thereof).
                 */
                flkp->l_type = F_UNLCK;
        } else {
                /*
                 * Found a conflicting lock.  Set the
                 * caller's *flkp with the info, first
                 * converting to the caller's whence.
                 */
                (void) convoff(vp, &flk0, flkp->l_whence, (offset_t)offset);
                *flkp = flk0;
        }

        return (0);
}

static int
nlm_frlock_setlk(struct nlm_host *hostp, vnode_t *vp,
    struct flock64 *flkp, int flags, u_offset_t offset,
    struct netobj *fhp, struct flk_callback *flcb,
    int vers, bool_t do_block)
{
        int error, xflags;

        error = convoff(vp, flkp, 0, (offset_t)offset);
        if (error != 0)
                return (error);

        /*
         * NFS v2 clients should not request locks where any part
         * of the lock range is beyond 0xffffffff.  The NFS code
         * checks that (see nfs_frlock, flk_check_lock_data), but
         * as that's outside this module, let's check here too.
         * This check ensures that we will be able to convert this
         * lock request into 32-bit form without change, and that
         * (more importantly) when the granted call back arrives,
         * it's unchanged when converted back into 64-bit form.
         * If this lock range were to change in any way during
         * either of those conversions, the "granted" call back
         * from the NLM server would not find our sleeping lock.
         */
        if (vers < NLM4_VERS) {
                if (flkp->l_start > MAX_UOFF32 ||
                    flkp->l_start + flkp->l_len > MAX_UOFF32 + 1)
                        return (EINVAL);
        }

        /*
         * Fill in l_sysid for the local locking calls.
         * Also, let's not trust the caller's l_pid.
         */
        flkp->l_sysid = hostp->nh_sysid | LM_SYSID_CLIENT;
        flkp->l_pid = curproc->p_pid;

        if (flkp->l_type == F_UNLCK) {
                /*
                 * Purge local (cached) lock information first,
                 * then clear the remote lock.
                 */
                (void) nlm_local_setlk(vp, flkp, flags);
                error = nlm_call_unlock(flkp, hostp, fhp, vers);

                return (error);
        }

        if (!do_block) {
                /*
                 * This is a non-blocking "set" request,
                 * so we can check locally first, and
                 * sometimes avoid an RPC call.
                 */
                struct flock64 flk0;

                flk0 = *flkp;
                error = nlm_local_getlk(vp, &flk0, flags);
                if (error != 0 && flk0.l_type != F_UNLCK) {
                        /* Found a conflicting lock. */
                        return (EAGAIN);
                }

                xflags = 0;
        } else {
                xflags = NLM_X_BLOCKING;
        }

        nfs_add_locking_id(vp, curproc->p_pid, RLMPL_PID,
            (char *)&curproc->p_pid, sizeof (pid_t));

        error = nlm_call_lock(vp, flkp, hostp, fhp, flcb, vers, xflags);
        if (error != 0)
                return (error);

        /*
         * Save the lock locally.  This should not fail,
         * because the server is authoritative about locks
         * and it just told us we have the lock!
         */
        error = nlm_local_setlk(vp, flkp, flags);
        if (error != 0) {
                /*
                 * That's unexpected situation. Just ignore the error.
                 */
                NLM_WARN("nlm_frlock_setlk: Failed to set local lock. "
                    "[err=%d]\n", error);
                error = 0;
        }

        return (error);
}

/*
 * Cancel all client side remote locks/shares on the
 * given host. Report to the processes that own
 * cancelled locks that they are removed by force
 * by sending SIGLOST.
 */
void
nlm_client_cancel_all(struct nlm_globals *g, struct nlm_host *hostp)
{
        struct locklist *llp_head, *llp;
        struct nlm_shres *nsp_head, *nsp;
        struct netobj lm_fh;
        rpcvers_t vers;
        int error, sysid;

        sysid = hostp->nh_sysid | LM_SYSID_CLIENT;
        nlm_host_cancel_slocks(g, hostp);

        /*
         * Destroy all active locks
         */
        llp_head = llp = flk_get_active_locks(sysid, NOPID);
        while (llp != NULL) {
                llp->ll_flock.l_type = F_UNLCK;

                error = nlm_init_fh_by_vp(llp->ll_vp, &lm_fh, &vers);
                if (error == 0)
                        (void) nlm_call_unlock(&llp->ll_flock, hostp,
                            &lm_fh, vers);

                nlm_local_cancelk(llp->ll_vp, &llp->ll_flock);
                llp = llp->ll_next;
        }

        flk_free_locklist(llp_head);

        /*
         * Destroy all active share reservations
         */
        nsp_head = nsp = nlm_get_active_shres(hostp);
        while (nsp != NULL) {
                error = nlm_init_fh_by_vp(nsp->ns_vp, &lm_fh, &vers);
                if (error == 0)
                        (void) nlm_call_unshare(nsp->ns_shr, hostp,
                            &lm_fh, vers);

                nlm_local_shrcancel(nsp->ns_vp, nsp->ns_shr);
                nlm_shres_untrack(hostp, nsp->ns_vp, nsp->ns_shr);
                nsp = nsp->ns_next;
        }

        nlm_free_shrlist(nsp_head);
}

/*
 * The function determines whether the lock "fl" can
 * be safely applied to the file vnode "vp" corresponds to.
 * The lock can be "safely" applied if all the conditions
 * above are held:
 *  - It's not a mandatory lock
 *  - The vnode wasn't mapped by anyone
 *  - The vnode was mapped, but it hasn't any locks on it.
 *  - The vnode was mapped and all locks it has occupies
 *    the whole file.
 */
int
nlm_safelock(vnode_t *vp, const struct flock64 *fl, cred_t *cr)
{
        rnode_t *rp = VTOR(vp);
        struct vattr va;
        int err;

        if ((rp->r_mapcnt > 0) && (fl->l_start != 0 || fl->l_len != 0))
                return (0);

        va.va_mask = AT_MODE;
        err = VOP_GETATTR(vp, &va, 0, cr, NULL);
        if (err != 0)
                return (0);

        /* NLM4 doesn't allow mandatory file locking */
        if (MANDLOCK(vp, va.va_mode))
                return (0);

        return (1);
}

/*
 * The function determines whether it's safe to map
 * a file correspoding to vnode vp.
 * The mapping is considered to be "safe" if file
 * either has no any locks on it or all locks it
 * has occupy the whole file.
 */
int
nlm_safemap(const vnode_t *vp)
{
        struct locklist *llp, *llp_next;
        struct nlm_slock *nslp;
        struct nlm_globals *g;
        int safe = 1;

        /* Check active locks at first */
        llp = flk_active_locks_for_vp(vp);
        while (llp != NULL) {
                if ((llp->ll_vp == vp) &&
                    !NLM_FLOCK_IS_SAFE(&llp->ll_flock))
                        safe = 0;

                llp_next = llp->ll_next;
                VN_RELE(llp->ll_vp);
                kmem_free(llp, sizeof (*llp));
                llp = llp_next;
        }
        if (!safe)
                return (safe);

        /* Then check sleeping locks if any */
        g = zone_getspecific(nlm_zone_key, curzone);
        mutex_enter(&g->lock);
        TAILQ_FOREACH(nslp, &g->nlm_slocks, nsl_link) {
                if (nslp->nsl_state == NLM_SL_BLOCKED &&
                    nslp->nsl_vp == vp &&
                    (nslp->nsl_lock.l_offset != 0 ||
                    nslp->nsl_lock.l_len != 0)) {
                        safe = 0;
                        break;
                }
        }

        mutex_exit(&g->lock);
        return (safe);
}

int
nlm_has_sleep(const vnode_t *vp)
{
        struct nlm_globals *g;
        struct nlm_slock *nslp;
        int has_slocks = FALSE;

        g = zone_getspecific(nlm_zone_key, curzone);
        mutex_enter(&g->lock);
        TAILQ_FOREACH(nslp, &g->nlm_slocks, nsl_link) {
                if (nslp->nsl_state == NLM_SL_BLOCKED &&
                    nslp->nsl_vp == vp) {
                        has_slocks = TRUE;
                        break;
                }
        }

        mutex_exit(&g->lock);
        return (has_slocks);
}

void
nlm_register_lock_locally(struct vnode *vp, struct nlm_host *hostp,
    struct flock64 *flk, int flags, u_offset_t offset)
{
        int sysid = 0;

        if (hostp != NULL) {
                sysid = hostp->nh_sysid | LM_SYSID_CLIENT;
        }

        flk->l_sysid = sysid;
        (void) convoff(vp, flk, 0, (offset_t)offset);
        (void) nlm_local_setlk(vp, flk, flags);
}


/*
 * The BSD code had functions here to "reclaim" (destroy)
 * remote locks when a vnode is being forcibly destroyed.
 * We just keep vnodes around until statd tells us the
 * client has gone away.
 */

static int
nlm_reclaim_lock(struct nlm_host *hostp, vnode_t *vp,
    struct flock64 *flp, int32_t orig_state)
{
        struct netobj lm_fh;
        int error, state;
        rpcvers_t vers;

        /*
         * If the remote NSM state changes during recovery, the host
         * must have rebooted a second time. In that case, we must
         * restart the recovery.
         */
        state = nlm_host_get_state(hostp);
        if (state != orig_state)
                return (ERESTART);

        error = nlm_init_fh_by_vp(vp, &lm_fh, &vers);
        if (error != 0)
                return (error);

        return (nlm_call_lock(vp, flp, hostp, &lm_fh,
            NULL, vers, NLM_X_RECLAIM));
}

/*
 * Get local lock information for some NFS server.
 *
 * This gets (checks for) a local conflicting lock.
 * Note: Modifies passed flock, if a conflict is found,
 * but the caller expects that.
 */
static int
nlm_local_getlk(vnode_t *vp, struct flock64 *fl, int flags)
{
        VERIFY(fl->l_whence == SEEK_SET);
        return (reclock(vp, fl, 0, flags, 0, NULL));
}

/*
 * Set local lock information for some NFS server.
 *
 * Called after a lock request (set or clear) succeeded. We record the
 * details in the local lock manager. Note that since the remote
 * server has granted the lock, we can be sure that it doesn't
 * conflict with any other locks we have in the local lock manager.
 *
 * Since it is possible that host may also make NLM client requests to
 * our NLM server, we use a different sysid value to record our own
 * client locks.
 *
 * Note that since it is possible for us to receive replies from the
 * server in a different order than the locks were granted (e.g. if
 * many local threads are contending for the same lock), we must use a
 * blocking operation when registering with the local lock manager.
 * We expect that any actual wait will be rare and short hence we
 * ignore signals for this.
 */
static int
nlm_local_setlk(vnode_t *vp, struct flock64 *fl, int flags)
{
        VERIFY(fl->l_whence == SEEK_SET);
        return (reclock(vp, fl, SETFLCK, flags, 0, NULL));
}

/*
 * Cancel local lock and send send SIGLOST signal
 * to the lock owner.
 *
 * NOTE: modifies flp
 */
static void
nlm_local_cancelk(vnode_t *vp, struct flock64 *flp)
{
        flp->l_type = F_UNLCK;
        (void) nlm_local_setlk(vp, flp, FREAD | FWRITE);
        nlm_send_siglost(flp->l_pid);
}

/*
 * Do NLM_LOCK call.
 * Was: nlm_setlock()
 *
 * NOTE: nlm_call_lock() function should care about locking/unlocking
 * of rnode->r_lkserlock which should be released before nlm_call_lock()
 * sleeps on waiting lock and acquired when it wakes up.
 */
static int
nlm_call_lock(vnode_t *vp, struct flock64 *flp,
        struct nlm_host *hostp, struct netobj *fhp,
        struct flk_callback *flcb, int vers, int xflags)
{
        struct nlm4_lockargs args;
        struct nlm_owner_handle oh;
        struct nlm_globals *g;
        rnode_t *rnp = VTOR(vp);
        struct nlm_slock *nslp = NULL;
        uint32_t xid;
        int error = 0;

        bzero(&args, sizeof (args));
        g = zone_getspecific(nlm_zone_key, curzone);
        nlm_init_lock(&args.alock, flp, fhp, &oh);

        args.exclusive = (flp->l_type == F_WRLCK);
        args.reclaim = xflags & NLM_X_RECLAIM;
        args.state = g->nsm_state;
        args.cookie.n_len = sizeof (xid);
        args.cookie.n_bytes = (char *)&xid;

        oh.oh_sysid = hostp->nh_sysid;
        xid = atomic_inc_32_nv(&nlm_xid);

        if (xflags & NLM_X_BLOCKING) {
                args.block = TRUE;
                nslp = nlm_slock_register(g, hostp, &args.alock, vp);
        }

        for (;;) {
                nlm_rpc_t *rpcp;
                enum clnt_stat stat;
                struct nlm4_res res;
                enum nlm4_stats nlm_err;

                error = nlm_host_get_rpc(hostp, vers, &rpcp);
                if (error != 0) {
                        error = ENOLCK;
                        goto out;
                }

                bzero(&res, sizeof (res));
                stat = nlm_lock_rpc(&args, &res, rpcp->nr_handle, vers);
                nlm_host_rele_rpc(hostp, rpcp);

                error = nlm_map_clnt_stat(stat);
                if (error != 0) {
                        if (error == EAGAIN)
                                continue;

                        goto out;
                }

                DTRACE_PROBE1(lock__res, enum nlm4_stats, res.stat.stat);
                nlm_err = res.stat.stat;
                xdr_free((xdrproc_t)xdr_nlm4_res, (void *)&res);
                if (nlm_err == nlm4_denied_grace_period) {
                        if (args.reclaim) {
                                error = ENOLCK;
                                goto out;
                        }

                        error = nlm_host_wait_grace(hostp);
                        if (error != 0)
                                goto out;

                        continue;
                }

                switch (nlm_err) {
                case nlm4_granted:
                case nlm4_blocked:
                        error = 0;
                        break;

                case nlm4_denied:
                        if (nslp != NULL) {
                                NLM_WARN("nlm_call_lock: got nlm4_denied for "
                                    "blocking lock\n");
                        }

                        error = EAGAIN;
                        break;

                default:
                        error = nlm_map_status(nlm_err);
                }

                /*
                 * If we deal with either non-blocking lock or
                 * with a blocking locks that wasn't blocked on
                 * the server side (by some reason), our work
                 * is finished.
                 */
                if (nslp == NULL                        ||
                    nlm_err != nlm4_blocked             ||
                    error != 0)
                        goto out;

                /*
                 * Before releasing the r_lkserlock of rnode, we should
                 * check whether the new lock is "safe". If it's not
                 * safe, disable caching for the given vnode. That is done
                 * for sleeping locks only that are waiting for a GRANT reply
                 * from the NLM server.
                 *
                 * NOTE: the vnode cache can be enabled back later if an
                 * unsafe lock will be merged with existent locks so that
                 * it will become safe. This condition is checked in the
                 * NFSv3 code (see nfs_lockcompletion).
                 */
                if (!NLM_FLOCK_IS_SAFE(flp)) {
                        mutex_enter(&vp->v_lock);
                        vp->v_flag &= ~VNOCACHE;
                        mutex_exit(&vp->v_lock);
                }

                /*
                 * The server should call us back with a
                 * granted message when the lock succeeds.
                 * In order to deal with broken servers,
                 * lost granted messages, or server reboots,
                 * we will also re-try every few seconds.
                 *
                 * Note: We're supposed to call these
                 * flk_invoke_callbacks when blocking.
                 * Take care on rnode->r_lkserlock, we should
                 * release it before going to sleep.
                 */
                (void) flk_invoke_callbacks(flcb, FLK_BEFORE_SLEEP);
                nfs_rw_exit(&rnp->r_lkserlock);

                error = nlm_slock_wait(g, nslp, g->retrans_tmo);

                /*
                 * NFS expects that we return with rnode->r_lkserlock
                 * locked on write, lock it back.
                 *
                 * NOTE: nfs_rw_enter_sig() can be either interruptible
                 * or not. It depends on options of NFS mount. Here
                 * we're _always_ uninterruptible (independently of mount
                 * options), because nfs_frlock/nfs3_frlock expects that
                 * we return with rnode->r_lkserlock acquired. So we don't
                 * want our lock attempt to be interrupted by a signal.
                 */
                (void) nfs_rw_enter_sig(&rnp->r_lkserlock, RW_WRITER, 0);
                (void) flk_invoke_callbacks(flcb, FLK_AFTER_SLEEP);

                if (error == 0) {
                        break;
                } else if (error == EINTR) {
                        /*
                         * We need to call the server to cancel our
                         * lock request.
                         */
                        DTRACE_PROBE1(cancel__lock, int, error);
                        (void) nlm_call_cancel(&args, hostp, vers);
                        break;
                } else {
                        /*
                         * Timeout happened, resend the lock request to
                         * the server. Well, we're a bit paranoid here,
                         * but keep in mind previous request could lost
                         * (especially with conectionless transport).
                         */

                        ASSERT(error == ETIMEDOUT);
                        continue;
                }
        }

        /*
         * We could disable the vnode cache for the given _sleeping_
         * (codition: nslp != NULL) lock if it was unsafe. Normally,
         * nfs_lockcompletion() function can enable the vnode cache
         * back if the lock becomes safe after activativation. But it
         * will not happen if any error occurs on the locking path.
         *
         * Here we enable the vnode cache back if the error occurred
         * and if there aren't any unsafe locks on the given vnode.
         * Note that if error happened, sleeping lock was derigistered.
         */
        if (error != 0 && nslp != NULL && nlm_safemap(vp)) {
                mutex_enter(&vp->v_lock);
                vp->v_flag |= VNOCACHE;
                mutex_exit(&vp->v_lock);
        }

out:
        if (nslp != NULL)
                nlm_slock_unregister(g, nslp);

        return (error);
}

/*
 * Do NLM_CANCEL call.
 * Helper for nlm_call_lock() error recovery.
 */
static int
nlm_call_cancel(struct nlm4_lockargs *largs,
        struct nlm_host *hostp, int vers)
{
        nlm4_cancargs cargs;
        uint32_t xid;
        int error, retries;

        bzero(&cargs, sizeof (cargs));

        xid = atomic_inc_32_nv(&nlm_xid);
        cargs.cookie.n_len = sizeof (xid);
        cargs.cookie.n_bytes = (char *)&xid;
        cargs.block     = largs->block;
        cargs.exclusive = largs->exclusive;
        cargs.alock     = largs->alock;

        /*
         * Unlike all other nlm_call_* functions, nlm_call_cancel
         * doesn't spin forever until it gets reasonable response
         * from NLM server. It makes limited number of retries and
         * if server doesn't send a reasonable reply, it returns an
         * error. It behaves like that because it's called from nlm_call_lock
         * with blocked signals and thus it can not be interrupted from
         * user space.
         */
        for (retries = 0; retries < NLM_CANCEL_NRETRS; retries++) {
                nlm_rpc_t *rpcp;
                enum clnt_stat stat;
                struct nlm4_res res;

                error = nlm_host_get_rpc(hostp, vers, &rpcp);
                if (error != 0)
                        return (ENOLCK);

                bzero(&res, sizeof (res));
                stat = nlm_cancel_rpc(&cargs, &res, rpcp->nr_handle, vers);
                nlm_host_rele_rpc(hostp, rpcp);

                DTRACE_PROBE1(cancel__rloop_end, enum clnt_stat, stat);
                error = nlm_map_clnt_stat(stat);
                if (error != 0) {
                        if (error == EAGAIN)
                                continue;

                        return (error);
                }

                DTRACE_PROBE1(cancel__res, enum nlm4_stats, res.stat.stat);
                switch (res.stat.stat) {
                        /*
                         * There was nothing to cancel. We are going to go ahead
                         * and assume we got the lock.
                         */
                case nlm_denied:
                        /*
                         * The server has recently rebooted.  Treat this as a
                         * successful cancellation.
                         */
                case nlm4_denied_grace_period:
                        /*
                         * We managed to cancel.
                         */
                case nlm4_granted:
                        error = 0;
                        break;

                default:
                        /*
                         * Broken server implementation.  Can't really do
                         * anything here.
                         */
                        error = EIO;
                        break;
                }

                xdr_free((xdrproc_t)xdr_nlm4_res, (void *)&res);
                break;
        }

        return (error);
}

/*
 * Do NLM_UNLOCK call.
 * Was: nlm_clearlock
 */
static int
nlm_call_unlock(struct flock64 *flp, struct nlm_host *hostp,
    struct netobj *fhp, int vers)
{
        struct nlm4_unlockargs args;
        struct nlm_owner_handle oh;
        enum nlm4_stats nlm_err;
        uint32_t xid;
        int error;

        bzero(&args, sizeof (args));
        nlm_init_lock(&args.alock, flp, fhp, &oh);

        oh.oh_sysid = hostp->nh_sysid;
        xid = atomic_inc_32_nv(&nlm_xid);
        args.cookie.n_len = sizeof (xid);
        args.cookie.n_bytes = (char *)&xid;

        for (;;) {
                nlm_rpc_t *rpcp;
                struct nlm4_res res;
                enum clnt_stat stat;

                error = nlm_host_get_rpc(hostp, vers, &rpcp);
                if (error != 0)
                        return (ENOLCK);

                bzero(&res, sizeof (res));
                stat = nlm_unlock_rpc(&args, &res, rpcp->nr_handle, vers);
                nlm_host_rele_rpc(hostp, rpcp);

                error = nlm_map_clnt_stat(stat);
                if (error != 0) {
                        if (error == EAGAIN)
                                continue;

                        return (error);
                }

                DTRACE_PROBE1(unlock__res, enum nlm4_stats, res.stat.stat);
                nlm_err = res.stat.stat;
                xdr_free((xdrproc_t)xdr_nlm4_res, (void *)&res);
                if (nlm_err == nlm4_denied_grace_period) {
                        error = nlm_host_wait_grace(hostp);
                        if (error != 0)
                                return (error);

                        continue;
                }

                break;
        }

        /* special cases */
        switch (nlm_err) {
        case nlm4_denied:
                error = EINVAL;
                break;
        default:
                error = nlm_map_status(nlm_err);
                break;
        }

        return (error);
}

/*
 * Do NLM_TEST call.
 * Was: nlm_getlock()
 */
static int
nlm_call_test(struct flock64 *flp, struct nlm_host *hostp,
    struct netobj *fhp, int vers)
{
        struct nlm4_testargs args;
        struct nlm4_holder h;
        struct nlm_owner_handle oh;
        enum nlm4_stats nlm_err;
        uint32_t xid;
        int error;

        bzero(&args, sizeof (args));
        nlm_init_lock(&args.alock, flp, fhp, &oh);

        args.exclusive = (flp->l_type == F_WRLCK);
        oh.oh_sysid = hostp->nh_sysid;
        xid = atomic_inc_32_nv(&nlm_xid);
        args.cookie.n_len = sizeof (xid);
        args.cookie.n_bytes = (char *)&xid;

        for (;;) {
                nlm_rpc_t *rpcp;
                struct nlm4_testres res;
                enum clnt_stat stat;

                error = nlm_host_get_rpc(hostp, vers, &rpcp);
                if (error != 0)
                        return (ENOLCK);

                bzero(&res, sizeof (res));
                stat = nlm_test_rpc(&args, &res, rpcp->nr_handle, vers);
                nlm_host_rele_rpc(hostp, rpcp);

                error = nlm_map_clnt_stat(stat);
                if (error != 0) {
                        if (error == EAGAIN)
                                continue;

                        return (error);
                }

                DTRACE_PROBE1(test__res, enum nlm4_stats, res.stat.stat);
                nlm_err = res.stat.stat;
                bcopy(&res.stat.nlm4_testrply_u.holder, &h, sizeof (h));
                xdr_free((xdrproc_t)xdr_nlm4_testres, (void *)&res);
                if (nlm_err == nlm4_denied_grace_period) {
                        error = nlm_host_wait_grace(hostp);
                        if (error != 0)
                                return (error);

                        continue;
                }

                break;
        }

        switch (nlm_err) {
        case nlm4_granted:
                flp->l_type = F_UNLCK;
                error = 0;
                break;

        case nlm4_denied:
                flp->l_start = h.l_offset;
                flp->l_len = h.l_len;
                flp->l_pid = h.svid;
                flp->l_type = (h.exclusive) ? F_WRLCK : F_RDLCK;
                flp->l_whence = SEEK_SET;
                flp->l_sysid = 0;
                error = 0;
                break;

        default:
                error = nlm_map_status(nlm_err);
                break;
        }

        return (error);
}


static void
nlm_init_lock(struct nlm4_lock *lock,
        const struct flock64 *fl, struct netobj *fh,
        struct nlm_owner_handle *oh)
{

        /* Caller converts to zero-base. */
        VERIFY(fl->l_whence == SEEK_SET);
        bzero(lock, sizeof (*lock));
        bzero(oh, sizeof (*oh));

        lock->caller_name = uts_nodename();
        lock->fh.n_len = fh->n_len;
        lock->fh.n_bytes = fh->n_bytes;
        lock->oh.n_len = sizeof (*oh);
        lock->oh.n_bytes = (void *)oh;
        lock->svid = fl->l_pid;
        lock->l_offset = fl->l_start;
        lock->l_len = fl->l_len;
}

/* ************************************************************** */

int
nlm_shrlock(struct vnode *vp, int cmd, struct shrlock *shr,
        int flags, struct netobj *fh, int vers)
{
        struct shrlock shlk;
        mntinfo_t *mi;
        servinfo_t *sv;
        const char *netid;
        struct nlm_host *host = NULL;
        int error;
        struct nlm_globals *g;

        mi = VTOMI(vp);
        sv = mi->mi_curr_serv;

        netid = nlm_knc_to_netid(sv->sv_knconf);
        if (netid == NULL) {
                NLM_ERR("nlm_shrlock: unknown NFS netid\n");
                return (ENOSYS);
        }

        g = zone_getspecific(nlm_zone_key, curzone);
        host = nlm_host_findcreate(g, sv->sv_hostname, netid, &sv->sv_addr);
        if (host == NULL)
                return (ENOSYS);

        /*
         * Fill in s_sysid for the local locking calls.
         * Also, let's not trust the caller's l_pid.
         */
        shlk = *shr;
        shlk.s_sysid = host->nh_sysid | LM_SYSID_CLIENT;
        shlk.s_pid = curproc->p_pid;

        if (cmd == F_UNSHARE) {
                /*
                 * Purge local (cached) share information first,
                 * then clear the remote share.
                 */
                (void) nlm_local_shrlock(vp, &shlk, cmd, flags);
                nlm_shres_untrack(host, vp, &shlk);
                error = nlm_call_unshare(&shlk, host, fh, vers);
                goto out;
        }

        nfs_add_locking_id(vp, curproc->p_pid, RLMPL_OWNER,
            shr->s_owner, shr->s_own_len);

        error = nlm_call_share(&shlk, host, fh, vers, FALSE);
        if (error != 0)
                goto out;

        /*
         * Save the share locally.  This should not fail,
         * because the server is authoritative about shares
         * and it just told us we have the share reservation!
         */
        error = nlm_local_shrlock(vp, shr, cmd, flags);
        if (error != 0) {
                /*
                 * Oh oh, we really don't expect an error here.
                 */
                NLM_WARN("nlm_shrlock: set locally, err %d\n", error);
                error = 0;
        }

        nlm_shres_track(host, vp, &shlk);
        nlm_host_monitor(g, host, 0);

out:
        nlm_host_release(g, host);

        return (error);
}

static int
nlm_reclaim_share(struct nlm_host *hostp, vnode_t *vp,
    struct shrlock *shr, uint32_t orig_state)
{
        struct netobj lm_fh;
        int error, state;
        rpcvers_t vers;

        state = nlm_host_get_state(hostp);
        if (state != orig_state) {
                /*
                 * It seems that NLM server rebooted while
                 * we were busy with recovery.
                 */
                return (ERESTART);
        }

        error = nlm_init_fh_by_vp(vp, &lm_fh, &vers);
        if (error != 0)
                return (error);

        return (nlm_call_share(shr, hostp, &lm_fh, vers, 1));
}

/*
 * Set local share information for some NFS server.
 *
 * Called after a share request (set or clear) succeeded. We record
 * the details in the local lock manager. Note that since the remote
 * server has granted the share, we can be sure that it doesn't
 * conflict with any other shares we have in the local lock manager.
 *
 * Since it is possible that host may also make NLM client requests to
 * our NLM server, we use a different sysid value to record our own
 * client shares.
 */
int
nlm_local_shrlock(vnode_t *vp, struct shrlock *shr, int cmd, int flags)
{
        return (fs_shrlock(vp, cmd, shr, flags, CRED(), NULL));
}

static void
nlm_local_shrcancel(vnode_t *vp, struct shrlock *shr)
{
        (void) nlm_local_shrlock(vp, shr, F_UNSHARE, FREAD | FWRITE);
        nlm_send_siglost(shr->s_pid);
}

/*
 * Do NLM_SHARE call.
 * Was: nlm_setshare()
 */
static int
nlm_call_share(struct shrlock *shr, struct nlm_host *host,
    struct netobj *fh, int vers, int reclaim)
{
        struct nlm4_shareargs args;
        enum nlm4_stats nlm_err;
        uint32_t xid;
        int error;

        bzero(&args, sizeof (args));
        nlm_init_share(&args.share, shr, fh);

        args.reclaim = reclaim;
        xid = atomic_inc_32_nv(&nlm_xid);
        args.cookie.n_len = sizeof (xid);
        args.cookie.n_bytes = (char *)&xid;


        for (;;) {
                nlm_rpc_t *rpcp;
                struct nlm4_shareres res;
                enum clnt_stat stat;

                error = nlm_host_get_rpc(host, vers, &rpcp);
                if (error != 0)
                        return (ENOLCK);

                bzero(&res, sizeof (res));
                stat = nlm_share_rpc(&args, &res, rpcp->nr_handle, vers);
                nlm_host_rele_rpc(host, rpcp);

                error = nlm_map_clnt_stat(stat);
                if (error != 0) {
                        if (error == EAGAIN)
                                continue;

                        return (error);
                }

                DTRACE_PROBE1(share__res, enum nlm4_stats, res.stat);
                nlm_err = res.stat;
                xdr_free((xdrproc_t)xdr_nlm4_shareres, (void *)&res);
                if (nlm_err == nlm4_denied_grace_period) {
                        if (args.reclaim)
                                return (ENOLCK);

                        error = nlm_host_wait_grace(host);
                        if (error != 0)
                                return (error);

                        continue;
                }

                break;
        }

        switch (nlm_err) {
        case nlm4_granted:
                error = 0;
                break;
        case nlm4_blocked:
        case nlm4_denied:
                error = EAGAIN;
                break;
        case nlm4_denied_nolocks:
        case nlm4_deadlck:
                error = ENOLCK;
                break;
        default:
                error = EINVAL;
                break;
        }

        return (error);
}

/*
 * Do NLM_UNSHARE call.
 */
static int
nlm_call_unshare(struct shrlock *shr, struct nlm_host *host,
    struct netobj *fh, int vers)
{
        struct nlm4_shareargs args;
        enum nlm4_stats nlm_err;
        uint32_t xid;
        int error;

        bzero(&args, sizeof (args));
        nlm_init_share(&args.share, shr, fh);

        xid = atomic_inc_32_nv(&nlm_xid);
        args.cookie.n_len = sizeof (xid);
        args.cookie.n_bytes = (char *)&xid;

        for (;;) {
                nlm_rpc_t *rpcp;
                struct nlm4_shareres res;
                enum clnt_stat stat;

                error = nlm_host_get_rpc(host, vers, &rpcp);
                if (error != 0)
                        return (ENOLCK);

                bzero(&res, sizeof (res));
                stat = nlm_unshare_rpc(&args, &res, rpcp->nr_handle, vers);
                nlm_host_rele_rpc(host, rpcp);

                error = nlm_map_clnt_stat(stat);
                if (error != 0) {
                        if (error == EAGAIN)
                                continue;

                        return (error);
                }

                DTRACE_PROBE1(unshare__res, enum nlm4_stats, res.stat);
                nlm_err = res.stat;
                xdr_free((xdrproc_t)xdr_nlm4_res, (void *)&res);
                if (nlm_err == nlm4_denied_grace_period) {
                        error = nlm_host_wait_grace(host);
                        if (error != 0)
                                return (error);

                        continue;
                }

                break;
        }

        switch (nlm_err) {
        case nlm4_granted:
                error = 0;
                break;
        case nlm4_denied:
                error = EAGAIN;
                break;
        case nlm4_denied_nolocks:
                error = ENOLCK;
                break;
        default:
                error = EINVAL;
                break;
        }

        return (error);
}

static void
nlm_init_share(struct nlm4_share *args,
        const struct shrlock *shr, struct netobj *fh)
{

        bzero(args, sizeof (*args));

        args->caller_name = uts_nodename();
        args->fh.n_len = fh->n_len;
        args->fh.n_bytes = fh->n_bytes;
        args->oh.n_len = shr->s_own_len;
        args->oh.n_bytes = (void *)shr->s_owner;

        switch (shr->s_deny) {
        default:
        case F_NODNY:
                args->mode = fsm_DN;
                break;
        case F_RDDNY:
                args->mode = fsm_DR;
                break;
        case F_WRDNY:
                args->mode = fsm_DW;
                break;
        case F_RWDNY:
                args->mode = fsm_DRW;
                break;
        }

        switch (shr->s_access) {
        default:
        case 0: /* seen with F_UNSHARE */
                args->access = fsa_NONE;
                break;
        case F_RDACC:
                args->access = fsa_R;
                break;
        case F_WRACC:
                args->access = fsa_W;
                break;
        case F_RWACC:
                args->access = fsa_RW;
                break;
        }
}

/*
 * Initialize filehandle according to the version
 * of NFS vnode was created on. The version of
 * NLM that can be used with given NFS version
 * is saved to lm_vers.
 */
static int
nlm_init_fh_by_vp(vnode_t *vp, struct netobj *fh, rpcvers_t *lm_vers)
{
        mntinfo_t *mi = VTOMI(vp);

        /*
         * Too bad the NFS code doesn't just carry the FH
         * in a netobj or a netbuf.
         */
        switch (mi->mi_vers) {
        case NFS_V3:
                /* See nfs3_frlock() */
                *lm_vers = NLM4_VERS;
                fh->n_len = VTOFH3(vp)->fh3_length;
                fh->n_bytes = (char *)&(VTOFH3(vp)->fh3_u.data);
                break;

        case NFS_VERSION:
                /* See nfs_frlock() */
                *lm_vers = NLM_VERS;
                fh->n_len = sizeof (fhandle_t);
                /* LINTED E_BAD_PTR_CAST_ALIGN */
                fh->n_bytes = (char *)VTOFH(vp);
                break;
        default:
                return (ENOSYS);
        }

        return (0);
}

/*
 * Send SIGLOST to the process identified by pid.
 * NOTE: called when NLM decides to remove lock
 * or share reservation ownder by the process
 * by force.
 */
static void
nlm_send_siglost(pid_t pid)
{
        proc_t *p;

        mutex_enter(&pidlock);
        p = prfind(pid);
        if (p != NULL)
                psignal(p, SIGLOST);

        mutex_exit(&pidlock);
}

static int
nlm_map_clnt_stat(enum clnt_stat stat)
{
        switch (stat) {
        case RPC_SUCCESS:
                return (0);

        case RPC_TIMEDOUT:
        case RPC_PROGUNAVAIL:
                return (EAGAIN);

        case RPC_INTR:
                return (EINTR);

        default:
                return (EINVAL);
        }
}

static int
nlm_map_status(enum nlm4_stats stat)
{
        switch (stat) {
        case nlm4_granted:
                return (0);

        case nlm4_denied:
                return (EAGAIN);

        case nlm4_denied_nolocks:
                return (ENOLCK);

        case nlm4_blocked:
                return (EAGAIN);

        case nlm4_denied_grace_period:
                return (EAGAIN);

        case nlm4_deadlck:
                return (EDEADLK);

        case nlm4_rofs:
                return (EROFS);

        case nlm4_stale_fh:
                return (ESTALE);

        case nlm4_fbig:
                return (EFBIG);

        case nlm4_failed:
                return (EACCES);

        default:
                return (EINVAL);
        }
}