Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / lockd / host.c
diff --git a/kernel/fs/lockd/host.c b/kernel/fs/lockd/host.c
new file mode 100644 (file)
index 0000000..969d589
--- /dev/null
@@ -0,0 +1,675 @@
+/*
+ * linux/fs/lockd/host.c
+ *
+ * Management for NLM peer hosts. The nlm_host struct is shared
+ * between client and server implementation. The only reason to
+ * do so is to reduce code bloat.
+ *
+ * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
+ */
+
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/in.h>
+#include <linux/in6.h>
+#include <linux/sunrpc/clnt.h>
+#include <linux/sunrpc/addr.h>
+#include <linux/sunrpc/svc.h>
+#include <linux/lockd/lockd.h>
+#include <linux/mutex.h>
+
+#include <linux/sunrpc/svc_xprt.h>
+
+#include <net/ipv6.h>
+
+#include "netns.h"
+
+#define NLMDBG_FACILITY                NLMDBG_HOSTCACHE
+#define NLM_HOST_NRHASH                32
+#define NLM_HOST_REBIND                (60 * HZ)
+#define NLM_HOST_EXPIRE                (300 * HZ)
+#define NLM_HOST_COLLECT       (120 * HZ)
+
+static struct hlist_head       nlm_server_hosts[NLM_HOST_NRHASH];
+static struct hlist_head       nlm_client_hosts[NLM_HOST_NRHASH];
+
+#define for_each_host(host, chain, table) \
+       for ((chain) = (table); \
+            (chain) < (table) + NLM_HOST_NRHASH; ++(chain)) \
+               hlist_for_each_entry((host), (chain), h_hash)
+
+#define for_each_host_safe(host, next, chain, table) \
+       for ((chain) = (table); \
+            (chain) < (table) + NLM_HOST_NRHASH; ++(chain)) \
+               hlist_for_each_entry_safe((host), (next), \
+                                               (chain), h_hash)
+
+static unsigned long           nrhosts;
+static DEFINE_MUTEX(nlm_host_mutex);
+
+static void                    nlm_gc_hosts(struct net *net);
+
+struct nlm_lookup_host_info {
+       const int               server;         /* search for server|client */
+       const struct sockaddr   *sap;           /* address to search for */
+       const size_t            salen;          /* it's length */
+       const unsigned short    protocol;       /* transport to search for*/
+       const u32               version;        /* NLM version to search for */
+       const char              *hostname;      /* remote's hostname */
+       const size_t            hostname_len;   /* it's length */
+       const int               noresvport;     /* use non-priv port */
+       struct net              *net;           /* network namespace to bind */
+};
+
+/*
+ * Hash function must work well on big- and little-endian platforms
+ */
+static unsigned int __nlm_hash32(const __be32 n)
+{
+       unsigned int hash = (__force u32)n ^ ((__force u32)n >> 16);
+       return hash ^ (hash >> 8);
+}
+
+static unsigned int __nlm_hash_addr4(const struct sockaddr *sap)
+{
+       const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
+       return __nlm_hash32(sin->sin_addr.s_addr);
+}
+
+static unsigned int __nlm_hash_addr6(const struct sockaddr *sap)
+{
+       const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
+       const struct in6_addr addr = sin6->sin6_addr;
+       return __nlm_hash32(addr.s6_addr32[0]) ^
+              __nlm_hash32(addr.s6_addr32[1]) ^
+              __nlm_hash32(addr.s6_addr32[2]) ^
+              __nlm_hash32(addr.s6_addr32[3]);
+}
+
+static unsigned int nlm_hash_address(const struct sockaddr *sap)
+{
+       unsigned int hash;
+
+       switch (sap->sa_family) {
+       case AF_INET:
+               hash = __nlm_hash_addr4(sap);
+               break;
+       case AF_INET6:
+               hash = __nlm_hash_addr6(sap);
+               break;
+       default:
+               hash = 0;
+       }
+       return hash & (NLM_HOST_NRHASH - 1);
+}
+
+/*
+ * Allocate and initialize an nlm_host.  Common to both client and server.
+ */
+static struct nlm_host *nlm_alloc_host(struct nlm_lookup_host_info *ni,
+                                      struct nsm_handle *nsm)
+{
+       struct nlm_host *host = NULL;
+       unsigned long now = jiffies;
+
+       if (nsm != NULL)
+               atomic_inc(&nsm->sm_count);
+       else {
+               host = NULL;
+               nsm = nsm_get_handle(ni->sap, ni->salen,
+                                       ni->hostname, ni->hostname_len);
+               if (unlikely(nsm == NULL)) {
+                       dprintk("lockd: %s failed; no nsm handle\n",
+                               __func__);
+                       goto out;
+               }
+       }
+
+       host = kmalloc(sizeof(*host), GFP_KERNEL);
+       if (unlikely(host == NULL)) {
+               dprintk("lockd: %s failed; no memory\n", __func__);
+               nsm_release(nsm);
+               goto out;
+       }
+
+       memcpy(nlm_addr(host), ni->sap, ni->salen);
+       host->h_addrlen    = ni->salen;
+       rpc_set_port(nlm_addr(host), 0);
+       host->h_srcaddrlen = 0;
+
+       host->h_rpcclnt    = NULL;
+       host->h_name       = nsm->sm_name;
+       host->h_version    = ni->version;
+       host->h_proto      = ni->protocol;
+       host->h_reclaiming = 0;
+       host->h_server     = ni->server;
+       host->h_noresvport = ni->noresvport;
+       host->h_inuse      = 0;
+       init_waitqueue_head(&host->h_gracewait);
+       init_rwsem(&host->h_rwsem);
+       host->h_state      = 0;
+       host->h_nsmstate   = 0;
+       host->h_pidcount   = 0;
+       atomic_set(&host->h_count, 1);
+       mutex_init(&host->h_mutex);
+       host->h_nextrebind = now + NLM_HOST_REBIND;
+       host->h_expires    = now + NLM_HOST_EXPIRE;
+       INIT_LIST_HEAD(&host->h_lockowners);
+       spin_lock_init(&host->h_lock);
+       INIT_LIST_HEAD(&host->h_granted);
+       INIT_LIST_HEAD(&host->h_reclaim);
+       host->h_nsmhandle  = nsm;
+       host->h_addrbuf    = nsm->sm_addrbuf;
+       host->net          = ni->net;
+
+out:
+       return host;
+}
+
+/*
+ * Destroy an nlm_host and free associated resources
+ *
+ * Caller must hold nlm_host_mutex.
+ */
+static void nlm_destroy_host_locked(struct nlm_host *host)
+{
+       struct rpc_clnt *clnt;
+       struct lockd_net *ln = net_generic(host->net, lockd_net_id);
+
+       dprintk("lockd: destroy host %s\n", host->h_name);
+
+       hlist_del_init(&host->h_hash);
+
+       nsm_unmonitor(host);
+       nsm_release(host->h_nsmhandle);
+
+       clnt = host->h_rpcclnt;
+       if (clnt != NULL)
+               rpc_shutdown_client(clnt);
+       kfree(host);
+
+       ln->nrhosts--;
+       nrhosts--;
+}
+
+/**
+ * nlmclnt_lookup_host - Find an NLM host handle matching a remote server
+ * @sap: network address of server
+ * @salen: length of server address
+ * @protocol: transport protocol to use
+ * @version: NLM protocol version
+ * @hostname: '\0'-terminated hostname of server
+ * @noresvport: 1 if non-privileged port should be used
+ *
+ * Returns an nlm_host structure that matches the passed-in
+ * [server address, transport protocol, NLM version, server hostname].
+ * If one doesn't already exist in the host cache, a new handle is
+ * created and returned.
+ */
+struct nlm_host *nlmclnt_lookup_host(const struct sockaddr *sap,
+                                    const size_t salen,
+                                    const unsigned short protocol,
+                                    const u32 version,
+                                    const char *hostname,
+                                    int noresvport,
+                                    struct net *net)
+{
+       struct nlm_lookup_host_info ni = {
+               .server         = 0,
+               .sap            = sap,
+               .salen          = salen,
+               .protocol       = protocol,
+               .version        = version,
+               .hostname       = hostname,
+               .hostname_len   = strlen(hostname),
+               .noresvport     = noresvport,
+               .net            = net,
+       };
+       struct hlist_head *chain;
+       struct nlm_host *host;
+       struct nsm_handle *nsm = NULL;
+       struct lockd_net *ln = net_generic(net, lockd_net_id);
+
+       dprintk("lockd: %s(host='%s', vers=%u, proto=%s)\n", __func__,
+                       (hostname ? hostname : "<none>"), version,
+                       (protocol == IPPROTO_UDP ? "udp" : "tcp"));
+
+       mutex_lock(&nlm_host_mutex);
+
+       chain = &nlm_client_hosts[nlm_hash_address(sap)];
+       hlist_for_each_entry(host, chain, h_hash) {
+               if (host->net != net)
+                       continue;
+               if (!rpc_cmp_addr(nlm_addr(host), sap))
+                       continue;
+
+               /* Same address. Share an NSM handle if we already have one */
+               if (nsm == NULL)
+                       nsm = host->h_nsmhandle;
+
+               if (host->h_proto != protocol)
+                       continue;
+               if (host->h_version != version)
+                       continue;
+
+               nlm_get_host(host);
+               dprintk("lockd: %s found host %s (%s)\n", __func__,
+                       host->h_name, host->h_addrbuf);
+               goto out;
+       }
+
+       host = nlm_alloc_host(&ni, nsm);
+       if (unlikely(host == NULL))
+               goto out;
+
+       hlist_add_head(&host->h_hash, chain);
+       ln->nrhosts++;
+       nrhosts++;
+
+       dprintk("lockd: %s created host %s (%s)\n", __func__,
+               host->h_name, host->h_addrbuf);
+
+out:
+       mutex_unlock(&nlm_host_mutex);
+       return host;
+}
+
+/**
+ * nlmclnt_release_host - release client nlm_host
+ * @host: nlm_host to release
+ *
+ */
+void nlmclnt_release_host(struct nlm_host *host)
+{
+       if (host == NULL)
+               return;
+
+       dprintk("lockd: release client host %s\n", host->h_name);
+
+       WARN_ON_ONCE(host->h_server);
+
+       if (atomic_dec_and_test(&host->h_count)) {
+               WARN_ON_ONCE(!list_empty(&host->h_lockowners));
+               WARN_ON_ONCE(!list_empty(&host->h_granted));
+               WARN_ON_ONCE(!list_empty(&host->h_reclaim));
+
+               mutex_lock(&nlm_host_mutex);
+               nlm_destroy_host_locked(host);
+               mutex_unlock(&nlm_host_mutex);
+       }
+}
+
+/**
+ * nlmsvc_lookup_host - Find an NLM host handle matching a remote client
+ * @rqstp: incoming NLM request
+ * @hostname: name of client host
+ * @hostname_len: length of client hostname
+ *
+ * Returns an nlm_host structure that matches the [client address,
+ * transport protocol, NLM version, client hostname] of the passed-in
+ * NLM request.  If one doesn't already exist in the host cache, a
+ * new handle is created and returned.
+ *
+ * Before possibly creating a new nlm_host, construct a sockaddr
+ * for a specific source address in case the local system has
+ * multiple network addresses.  The family of the address in
+ * rq_daddr is guaranteed to be the same as the family of the
+ * address in rq_addr, so it's safe to use the same family for
+ * the source address.
+ */
+struct nlm_host *nlmsvc_lookup_host(const struct svc_rqst *rqstp,
+                                   const char *hostname,
+                                   const size_t hostname_len)
+{
+       struct hlist_head *chain;
+       struct nlm_host *host = NULL;
+       struct nsm_handle *nsm = NULL;
+       struct sockaddr *src_sap = svc_daddr(rqstp);
+       size_t src_len = rqstp->rq_daddrlen;
+       struct net *net = SVC_NET(rqstp);
+       struct nlm_lookup_host_info ni = {
+               .server         = 1,
+               .sap            = svc_addr(rqstp),
+               .salen          = rqstp->rq_addrlen,
+               .protocol       = rqstp->rq_prot,
+               .version        = rqstp->rq_vers,
+               .hostname       = hostname,
+               .hostname_len   = hostname_len,
+               .net            = net,
+       };
+       struct lockd_net *ln = net_generic(net, lockd_net_id);
+
+       dprintk("lockd: %s(host='%*s', vers=%u, proto=%s)\n", __func__,
+                       (int)hostname_len, hostname, rqstp->rq_vers,
+                       (rqstp->rq_prot == IPPROTO_UDP ? "udp" : "tcp"));
+
+       mutex_lock(&nlm_host_mutex);
+
+       if (time_after_eq(jiffies, ln->next_gc))
+               nlm_gc_hosts(net);
+
+       chain = &nlm_server_hosts[nlm_hash_address(ni.sap)];
+       hlist_for_each_entry(host, chain, h_hash) {
+               if (host->net != net)
+                       continue;
+               if (!rpc_cmp_addr(nlm_addr(host), ni.sap))
+                       continue;
+
+               /* Same address. Share an NSM handle if we already have one */
+               if (nsm == NULL)
+                       nsm = host->h_nsmhandle;
+
+               if (host->h_proto != ni.protocol)
+                       continue;
+               if (host->h_version != ni.version)
+                       continue;
+               if (!rpc_cmp_addr(nlm_srcaddr(host), src_sap))
+                       continue;
+
+               /* Move to head of hash chain. */
+               hlist_del(&host->h_hash);
+               hlist_add_head(&host->h_hash, chain);
+
+               nlm_get_host(host);
+               dprintk("lockd: %s found host %s (%s)\n",
+                       __func__, host->h_name, host->h_addrbuf);
+               goto out;
+       }
+
+       host = nlm_alloc_host(&ni, nsm);
+       if (unlikely(host == NULL))
+               goto out;
+
+       memcpy(nlm_srcaddr(host), src_sap, src_len);
+       host->h_srcaddrlen = src_len;
+       hlist_add_head(&host->h_hash, chain);
+       ln->nrhosts++;
+       nrhosts++;
+
+       dprintk("lockd: %s created host %s (%s)\n",
+               __func__, host->h_name, host->h_addrbuf);
+
+out:
+       mutex_unlock(&nlm_host_mutex);
+       return host;
+}
+
+/**
+ * nlmsvc_release_host - release server nlm_host
+ * @host: nlm_host to release
+ *
+ * Host is destroyed later in nlm_gc_host().
+ */
+void nlmsvc_release_host(struct nlm_host *host)
+{
+       if (host == NULL)
+               return;
+
+       dprintk("lockd: release server host %s\n", host->h_name);
+
+       WARN_ON_ONCE(!host->h_server);
+       atomic_dec(&host->h_count);
+}
+
+/*
+ * Create the NLM RPC client for an NLM peer
+ */
+struct rpc_clnt *
+nlm_bind_host(struct nlm_host *host)
+{
+       struct rpc_clnt *clnt;
+
+       dprintk("lockd: nlm_bind_host %s (%s)\n",
+                       host->h_name, host->h_addrbuf);
+
+       /* Lock host handle */
+       mutex_lock(&host->h_mutex);
+
+       /* If we've already created an RPC client, check whether
+        * RPC rebind is required
+        */
+       if ((clnt = host->h_rpcclnt) != NULL) {
+               if (time_after_eq(jiffies, host->h_nextrebind)) {
+                       rpc_force_rebind(clnt);
+                       host->h_nextrebind = jiffies + NLM_HOST_REBIND;
+                       dprintk("lockd: next rebind in %lu jiffies\n",
+                                       host->h_nextrebind - jiffies);
+               }
+       } else {
+               unsigned long increment = nlmsvc_timeout;
+               struct rpc_timeout timeparms = {
+                       .to_initval     = increment,
+                       .to_increment   = increment,
+                       .to_maxval      = increment * 6UL,
+                       .to_retries     = 5U,
+               };
+               struct rpc_create_args args = {
+                       .net            = host->net,
+                       .protocol       = host->h_proto,
+                       .address        = nlm_addr(host),
+                       .addrsize       = host->h_addrlen,
+                       .timeout        = &timeparms,
+                       .servername     = host->h_name,
+                       .program        = &nlm_program,
+                       .version        = host->h_version,
+                       .authflavor     = RPC_AUTH_UNIX,
+                       .flags          = (RPC_CLNT_CREATE_NOPING |
+                                          RPC_CLNT_CREATE_AUTOBIND),
+               };
+
+               /*
+                * lockd retries server side blocks automatically so we want
+                * those to be soft RPC calls. Client side calls need to be
+                * hard RPC tasks.
+                */
+               if (!host->h_server)
+                       args.flags |= RPC_CLNT_CREATE_HARDRTRY;
+               if (host->h_noresvport)
+                       args.flags |= RPC_CLNT_CREATE_NONPRIVPORT;
+               if (host->h_srcaddrlen)
+                       args.saddress = nlm_srcaddr(host);
+
+               clnt = rpc_create(&args);
+               if (!IS_ERR(clnt))
+                       host->h_rpcclnt = clnt;
+               else {
+                       printk("lockd: couldn't create RPC handle for %s\n", host->h_name);
+                       clnt = NULL;
+               }
+       }
+
+       mutex_unlock(&host->h_mutex);
+       return clnt;
+}
+
+/*
+ * Force a portmap lookup of the remote lockd port
+ */
+void
+nlm_rebind_host(struct nlm_host *host)
+{
+       dprintk("lockd: rebind host %s\n", host->h_name);
+       if (host->h_rpcclnt && time_after_eq(jiffies, host->h_nextrebind)) {
+               rpc_force_rebind(host->h_rpcclnt);
+               host->h_nextrebind = jiffies + NLM_HOST_REBIND;
+       }
+}
+
+/*
+ * Increment NLM host count
+ */
+struct nlm_host * nlm_get_host(struct nlm_host *host)
+{
+       if (host) {
+               dprintk("lockd: get host %s\n", host->h_name);
+               atomic_inc(&host->h_count);
+               host->h_expires = jiffies + NLM_HOST_EXPIRE;
+       }
+       return host;
+}
+
+static struct nlm_host *next_host_state(struct hlist_head *cache,
+                                       struct nsm_handle *nsm,
+                                       const struct nlm_reboot *info)
+{
+       struct nlm_host *host;
+       struct hlist_head *chain;
+
+       mutex_lock(&nlm_host_mutex);
+       for_each_host(host, chain, cache) {
+               if (host->h_nsmhandle == nsm
+                   && host->h_nsmstate != info->state) {
+                       host->h_nsmstate = info->state;
+                       host->h_state++;
+
+                       nlm_get_host(host);
+                       mutex_unlock(&nlm_host_mutex);
+                       return host;
+               }
+       }
+
+       mutex_unlock(&nlm_host_mutex);
+       return NULL;
+}
+
+/**
+ * nlm_host_rebooted - Release all resources held by rebooted host
+ * @info: pointer to decoded results of NLM_SM_NOTIFY call
+ *
+ * We were notified that the specified host has rebooted.  Release
+ * all resources held by that peer.
+ */
+void nlm_host_rebooted(const struct nlm_reboot *info)
+{
+       struct nsm_handle *nsm;
+       struct nlm_host *host;
+
+       nsm = nsm_reboot_lookup(info);
+       if (unlikely(nsm == NULL))
+               return;
+
+       /* Mark all hosts tied to this NSM state as having rebooted.
+        * We run the loop repeatedly, because we drop the host table
+        * lock for this.
+        * To avoid processing a host several times, we match the nsmstate.
+        */
+       while ((host = next_host_state(nlm_server_hosts, nsm, info)) != NULL) {
+               nlmsvc_free_host_resources(host);
+               nlmsvc_release_host(host);
+       }
+       while ((host = next_host_state(nlm_client_hosts, nsm, info)) != NULL) {
+               nlmclnt_recovery(host);
+               nlmclnt_release_host(host);
+       }
+
+       nsm_release(nsm);
+}
+
+static void nlm_complain_hosts(struct net *net)
+{
+       struct hlist_head *chain;
+       struct nlm_host *host;
+
+       if (net) {
+               struct lockd_net *ln = net_generic(net, lockd_net_id);
+
+               if (ln->nrhosts == 0)
+                       return;
+               printk(KERN_WARNING "lockd: couldn't shutdown host module for net %p!\n", net);
+               dprintk("lockd: %lu hosts left in net %p:\n", ln->nrhosts, net);
+       } else {
+               if (nrhosts == 0)
+                       return;
+               printk(KERN_WARNING "lockd: couldn't shutdown host module!\n");
+               dprintk("lockd: %lu hosts left:\n", nrhosts);
+       }
+
+       for_each_host(host, chain, nlm_server_hosts) {
+               if (net && host->net != net)
+                       continue;
+               dprintk("       %s (cnt %d use %d exp %ld net %p)\n",
+                       host->h_name, atomic_read(&host->h_count),
+                       host->h_inuse, host->h_expires, host->net);
+       }
+}
+
+void
+nlm_shutdown_hosts_net(struct net *net)
+{
+       struct hlist_head *chain;
+       struct nlm_host *host;
+
+       mutex_lock(&nlm_host_mutex);
+
+       /* First, make all hosts eligible for gc */
+       dprintk("lockd: nuking all hosts in net %p...\n", net);
+       for_each_host(host, chain, nlm_server_hosts) {
+               if (net && host->net != net)
+                       continue;
+               host->h_expires = jiffies - 1;
+               if (host->h_rpcclnt) {
+                       rpc_shutdown_client(host->h_rpcclnt);
+                       host->h_rpcclnt = NULL;
+               }
+       }
+
+       /* Then, perform a garbage collection pass */
+       nlm_gc_hosts(net);
+       mutex_unlock(&nlm_host_mutex);
+
+       nlm_complain_hosts(net);
+}
+
+/*
+ * Shut down the hosts module.
+ * Note that this routine is called only at server shutdown time.
+ */
+void
+nlm_shutdown_hosts(void)
+{
+       dprintk("lockd: shutting down host module\n");
+       nlm_shutdown_hosts_net(NULL);
+}
+
+/*
+ * Garbage collect any unused NLM hosts.
+ * This GC combines reference counting for async operations with
+ * mark & sweep for resources held by remote clients.
+ */
+static void
+nlm_gc_hosts(struct net *net)
+{
+       struct hlist_head *chain;
+       struct hlist_node *next;
+       struct nlm_host *host;
+
+       dprintk("lockd: host garbage collection for net %p\n", net);
+       for_each_host(host, chain, nlm_server_hosts) {
+               if (net && host->net != net)
+                       continue;
+               host->h_inuse = 0;
+       }
+
+       /* Mark all hosts that hold locks, blocks or shares */
+       nlmsvc_mark_resources(net);
+
+       for_each_host_safe(host, next, chain, nlm_server_hosts) {
+               if (net && host->net != net)
+                       continue;
+               if (atomic_read(&host->h_count) || host->h_inuse
+                || time_before(jiffies, host->h_expires)) {
+                       dprintk("nlm_gc_hosts skipping %s "
+                               "(cnt %d use %d exp %ld net %p)\n",
+                               host->h_name, atomic_read(&host->h_count),
+                               host->h_inuse, host->h_expires, host->net);
+                       continue;
+               }
+               nlm_destroy_host_locked(host);
+       }
+
+       if (net) {
+               struct lockd_net *ln = net_generic(net, lockd_net_id);
+
+               ln->next_gc = jiffies + NLM_HOST_COLLECT;
+       }
+}